Upload folder using huggingface_hub

icon_generation/backup/accepted_domains.txt

@@ -0,0 +1,346 @@
+Industry Sector
+Web Browser
+Entertainment Genre
+Media Format
+Smart Home Device Category
+Art Medium
+Product Category
+Land Use Type
+Device Type
+Crop Type
+Transportation Mode
+Smartphone Brand
+Vehicle Type
+Energy Technologies
+Energy Source
+Geographic Region
+Academic Subject
+Material
+Product Brand
+Client Engagement Channel
+Transportation Infrastructure Type
+Physical Activity
+Pet Category
+Clothing Type
+Dance Discipline
+Musical Instruments
+Building Type
+Household Appliance
+Incident Type
+Occupation
+Cuisine
+Season (Calendar & Climatic)
+Media Asset
+Company Name
+Live Performance Type
+Animal Species
+Event Type
+Mobile Phone Type
+Accommodation Type
+Operating System
+Vehicle Powertrain Type
+Streaming Service
+Weather Type
+Cooking Technique
+Religious Affiliation
+Sentiment
+Biome Type
+Playground Feature
+Handicraft Category
+Retail Location Type
+Art and Craft Techniques
+Park Type
+Camera Form Factor
+Residential Room Type
+Business Segment (Food & Beverage)
+Leisure Activity
+Museum Object Type
+Pollution Source or Pollutant Type
+Medical Injury Type
+Dietary Preference or Restriction
+Dietary Protein Source
+Coffee Brewing Method
+Social Event Type
+Purchase Channel
+Investment Asset Class
+Expense Category
+Music Genre
+Dwelling Type
+Time of Day
+Produce (Fruit & Vegetable)
+Medical Procedure
+Lighting Type
+Landmark Name
+Seafood Species
+Travel Category
+Color
+Cryptocurrency Name
+Food Category
+Fashion Accessory Category
+Tourism Type
+Interest Category
+Financial Incentive Type
+Affectionate Gestures
+Mobile OS
+Medical Specialty
+Health Topic
+Eating Occasion
+Storage Type
+Coffee Beverage / Preparation Style
+Environmental Impact Categories
+Game Genre
+Holiday Name
+Museum Focus
+Waste Material Type
+Freight Cargo Type
+Funding Source
+Civil Engineering Structure Type
+Recipe Ingredient
+Software Application Name
+Sofa Type
+Sales Offer Type
+Financial Product Type
+Festival Theme
+Camera Lens Type
+Concession Stand Product
+Climate Classification
+Primate Species
+Sports Facility Type
+Finishing Position
+Sustainability Category
+Audio Listening Method
+Travel Market Segment
+Fitness Goal
+Communication Channel
+Tourist Attraction
+Grain Type
+Furniture Category
+Heritage Site Type
+Beverage Type
+Research Publication Type
+Earring Type
+Common Illicit and Recreational Drugs
+Occasion (Usage Scenario)
+Major Animal Groups
+Alcoholic Beverage Type
+Establishment Type
+Website Content Section
+Property Flooring Type
+Satellite Mission Type
+Vehicle Propulsion Type
+Retail Store Category
+Fantasy Creature or Race
+Milestone Type
+Cyberattack Type
+Art Subject
+Citrus Variety
+Food Item
+Insect Common Name
+Insurance Type
+Venue Seating Section
+Home Decor Category
+Driving Environment
+Wearable Accessory Type
+Play Activity Type
+Music Distribution Format
+Sports Court Surface
+Application Category (App Store)
+Exercise Type
+Streaming Platforms
+Cultivation Environment
+Sensor Type
+Wine Grape Varieties and Styles
+Email Service Provider
+Costume Character
+Livestock Species
+Historical Period
+Class Type
+Religious Symbol
+Fruit Name
+Pollution Source
+Dietary Food Group
+Neighborhood Name
+Manufacturing Step
+Amusement Ride Type
+Dish Type
+Retailer Name
+Pesticide Type (Target/Function)
+Vehicle Maintenance Type
+Painting Medium
+Craft Supplies
+Project Phase
+Dominant Forest Cover Type
+Device Capabilities (Sensors, Connectivity & Features)
+Competition Gender Division
+Irrigation Source Type
+Gift Category
+Vessel Type
+Notable Sacred Sites
+Online Video Platforms
+Tropical Cyclone Intensity Category (Saffir–Simpson and related classifications)
+Internet Connectivity Status
+Water Supply Type
+Performance Terrain Type
+Workout Modality
+Common Pest Type
+Patrol Method
+Sustainable Building Feature
+Fatal Incident Type
+Learning Resource Type
+Service Category
+Manufacturing Process
+Pest Control Technique
+Personal Financial Concern Category
+Software Application
+Result Status (Test/Operation)
+Weightlifting and Strength Training Exercises
+Access Level
+Therapeutic Area
+Employment Type
+Event Category
+Wellness Program Type
+Packaging Material
+ADAS Feature
+Military Platform Type
+Subscription Plan Tier
+Organization Sector
+Planetary Rover Name
+Project Category
+Footwear Style
+Commercial Aircraft Model
+Forage Type
+Clinical Service Type
+Hat Style
+U.S. Military Service Branch
+App Permission
+News Organization
+Ranching Practices and Systems
+Construction & Infrastructure Project Type
+Home Feature
+Production Method
+Pricing Basis
+Booking Channel
+Educational Focus Area
+Ecosystem / Habitat Type
+Craft Materials
+Subsystem / Component Type
+Flavor Profile (Tasting Descriptors)
+Transport Route Type
+Basic Human Needs
+Fashion Style
+Humanitarian Aid Sector
+Character Type
+Aircraft Category (type/market segment)
+Technology Solution Type
+Roofing Material/Type
+Common Amphibian Species (common names)
+Mammal Species Name
+Insect Group (common names)
+Movie Theater Format
+Food Item Name
+Medical Treatment Type
+Handbag Style
+Artwork Title
+Personal Protective Equipment (PPE) Type
+Academic Subject Category
+Microphone Type
+Pollinator Types
+Computer Peripherals
+Agricultural Activities
+Mission Phase
+Certification Level
+Leisure Amenities
+Personal Relationship Type
+Astronomical Observatory Name
+Cultural Institution Type
+Historic Site Name
+Medical Interventions
+Loyalty Program Feature
+Racing Circuit Type
+Competitor Brand (Fashion & Apparel)
+Hazard Type
+Pet Service Type
+Pet Wellness Package Type
+Benefit Recipient Type
+Forms of Folklore
+Generational Cohort
+Vehicle Part Type
+Ad Format
+Jewelry-making Technique
+Cultural Attire
+Chinese New Year Foods
+Maize (Corn) Cultivar/Variety
+Consumer Electronics & Computer Hardware Category
+Educational Program Type
+Art Installation Type
+Tour Activity
+Server Role
+Emotion Type
+Cause of Damage
+Passenger Persona (Travel Behavior)
+Solar System Planet
+Healthcare Facility Type
+Application Functionality
+Role-Playing Game System
+Freight Transport Mode
+Product Variant (Production Method)
+Personal Data Type
+Financial Transaction Type
+Spending Occasion
+Home Improvement Project Type
+Cultural Art Traditions
+Ticket Category
+Hardware Component Type
+Manufacturing Machine Type
+Motorcycle Type
+Light Color (illumination)
+Endorsement Category (Products & Services)
+Communication Purpose
+Motor Vehicle Collision Type
+Cultural Offering Type
+Donor Classification
+Lemur Taxon Name (Genus and Species)
+Therapy Modality (Physical, Psychological, and Complementary Therapies)
+Body Shape (Figure Type)
+Comic Book Series
+Outdoor Lighting Fixture Type
+Material Sourcing Type
+Power Plant
+Driving Scenario
+Skilled Trade
+Land Use / Development Type
+Ritual Type
+High-Risk Patient Groups
+Subject Area (Topic)
+Exhibit Theme
+Venue Type
+Toy Type
+Harvesting Method
+Seabird Species
+Hazard Mitigation Project Type
+Accessibility Feature Type
+Tractor Powertrain & Control Configuration
+Filtration Technology
+Running Shoe Segment
+Art & Craft Workshop Discipline
+Scientific Discovery Category
+Monetization Method
+Debris Material
+Fishing Capture Method
+Land Management Practice
+Basketball Shot Zone
+Dress Silhouette
+Water Conservation Measure
+Love Language Type
+Crop Pest
+Gift Type
+Farm Diversification Strategies
+Payment Method
+Climate Zone (major types — Köppen & common names)
+Architectural Style
+Schedule Disruption Reason
+Tomato Variety
+Lighting Fixture Type
+Washing Machine Type
+Art Supply Set Type
+Gender Identity

icon_generation/backup/analyze_domains.py

@@ -0,0 +1,169 @@
+#!/usr/bin/env python3
+"""
+分析 domain_value_pairs_filtered.txt
+提取所有唯一的 domain 并统计信息
+"""
+
+from collections import defaultdict
+from typing import List, Dict, Tuple
+
+
+def parse_csv_line(line: str) -> Tuple[str, str, int]:
+    """
+    解析 CSV 行，处理可能包含逗号的带引号字段
+    
+    Returns:
+        (domain, specific_attribute, count)
+    """
+    parts = []
+    current = []
+    in_quotes = False
+    
+    for char in line:
+        if char == '"':
+            in_quotes = not in_quotes
+        elif char == ',' and not in_quotes:
+            parts.append(''.join(current).strip())
+            current = []
+        else:
+            current.append(char)
+    parts.append(''.join(current).strip())
+    
+    if len(parts) >= 3:
+        domain = parts[0]
+        attribute = parts[1]
+        count = int(parts[2])
+        return domain, attribute, count
+    else:
+        return None, None, 0
+
+
+def analyze_domains(file_path: str):
+    """分析文件并提取 domain 信息"""
+    
+    print("=" * 80)
+    print("Domain 分析工具")
+    print("=" * 80)
+    print()
+    
+    print(f"📖 读取文件: {file_path}")
+    
+    # 统计信息
+    domain_stats = defaultdict(lambda: {
+        'count': 0,
+        'total_value_count': 0,
+        'attributes': [],
+        'top_attributes': []
+    })
+    
+    total_lines = 0
+    
+    # 读取文件
+    with open(file_path, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                continue
+            
+            domain, attribute, count = parse_csv_line(line)
+            if domain:
+                total_lines += 1
+                domain_stats[domain]['count'] += 1
+                domain_stats[domain]['total_value_count'] += count
+                domain_stats[domain]['attributes'].append((attribute, count))
+    
+    print(f"✅ 成功读取 {total_lines} 行数据")
+    print()
+    
+    # 排序 domains（按 total_value_count 降序）
+    sorted_domains = sorted(
+        domain_stats.items(),
+        key=lambda x: x[1]['total_value_count'],
+        reverse=True
+    )
+    
+    # 输出统计信息
+    print("=" * 80)
+    print("📊 统计信息:")
+    print("=" * 80)
+    print(f"唯一 Domain 数量: {len(sorted_domains)}")
+    print(f"总 Pairs 数量: {total_lines}")
+    print(f"总 Count 数: {sum(stats['total_value_count'] for _, stats in sorted_domains):,}")
+    print()
+    
+    # 输出所有 domains
+    print("=" * 80)
+    print("📋 所有 Domains 列表 (按 total_count 降序):")
+    print("=" * 80)
+    print()
+    
+    for idx, (domain, stats) in enumerate(sorted_domains, 1):
+        num_attributes = stats['count']
+        total_count = stats['total_value_count']
+        
+        # 获取 top 3 attributes
+        top_attrs = sorted(stats['attributes'], key=lambda x: x[1], reverse=True)[:3]
+        top_attrs_str = ', '.join([f"{attr} ({cnt:,})" for attr, cnt in top_attrs])
+        
+        print(f"{idx:3d}. {domain}")
+        print(f"     - Attributes 数量: {num_attributes}")
+        print(f"     - 总 Count: {total_count:,}")
+        print(f"     - Top Attributes: {top_attrs_str}")
+        print()
+    
+    print("=" * 80)
+    
+    # 输出简洁的 domain 列表
+    print("\n📝 简洁 Domain 列表:")
+    print("=" * 80)
+    domain_list = [domain for domain, _ in sorted_domains]
+    for i in range(0, len(domain_list), 3):
+        line_domains = domain_list[i:i+3]
+        print("  " + " | ".join(f"{d:30s}" for d in line_domains))
+    
+    print()
+    print("=" * 80)
+    
+    # 保存到文件
+    output_file = 'domains_list.txt'
+    with open(output_file, 'w', encoding='utf-8') as f:
+        f.write("所有 Domains 列表 (按 total_count 降序)\n")
+        f.write("=" * 80 + "\n\n")
+        
+        for idx, (domain, stats) in enumerate(sorted_domains, 1):
+            f.write(f"{idx}. {domain}\n")
+            f.write(f"   Attributes: {stats['count']}, Total Count: {stats['total_value_count']:,}\n")
+            
+            # Top 5 attributes
+            top_attrs = sorted(stats['attributes'], key=lambda x: x[1], reverse=True)[:5]
+            f.write(f"   Top Attributes:\n")
+            for attr, cnt in top_attrs:
+                f.write(f"      - {attr}: {cnt:,}\n")
+            f.write("\n")
+    
+    print(f"💾 详细信息已保存到: {output_file}")
+    
+    # 输出 Python 列表格式
+    print("\n🐍 Python 列表格式:")
+    print("=" * 80)
+    print("domains = [")
+    for domain in domain_list:
+        print(f"    '{domain}',")
+    print("]")
+    
+    return domain_list, domain_stats
+
+
+def main():
+    file_path = 'domain_value_pairs_filtered.txt'
+    domains, stats = analyze_domains(file_path)
+    
+    print()
+    print("=" * 80)
+    print("✨ 分析完成！")
+    print("=" * 80)
+
+
+if __name__ == '__main__':
+    main()
+

icon_generation/backup/check_progress.sh

@@ -0,0 +1,38 @@
+#!/bin/bash
+# 检查 refine_domains.py 的处理进度
+
+echo "========================================="
+echo "Domain Refinement 进度检查"
+echo "========================================="
+echo ""
+
+# 检查进程是否还在运行
+if pgrep -f "refine_domains.py" > /dev/null; then
+    echo "✅ 脚本正在运行中..."
+else
+    echo "⚠️  脚本未运行"
+fi
+
+echo ""
+echo "最新处理进度："
+echo "========================================="
+tail -30 /home/lizhen/ChartPipeline/icon_generation/refine_log.txt | grep -E "(处理进度|已用时间|当前字段|Value filtering 完成)"
+
+echo ""
+echo "========================================="
+echo "临时文件信息："
+if [ -f "/home/lizhen/ChartPipeline/icon_generation/refined_domains_temp.json" ]; then
+    temp_size=$(wc -l < /home/lizhen/ChartPipeline/icon_generation/refined_domains_temp.json)
+    echo "📄 临时文件行数: $temp_size"
+    
+    # 统计已处理的字段数
+    processed=$(grep -c '"domain":' /home/lizhen/ChartPipeline/icon_generation/refined_domains_temp.json)
+    echo "✅ 已处理字段数: $processed"
+else
+    echo "⚠️  临时文件不存在"
+fi
+
+echo ""
+echo "最后更新时间："
+ls -lh /home/lizhen/ChartPipeline/icon_generation/refine_log.txt | awk '{print $6, $7, $8}'
+

icon_generation/backup/domain_summary.txt

@@ -0,0 +1,514 @@
+Industry Sector
+Web Browser
+Entertainment Genre
+US State
+Digital Platform or Channel
+Media Format
+Smart Home Device Category
+Art Medium
+Product Category
+Continent
+Land Use Type
+Device Type
+Crop Type
+Transportation Mode
+Major Professional Sports Leagues (Global)
+Political Party
+Smartphone Brand
+Vehicle Type
+Energy Technologies
+Dog Breed
+Energy Source
+Irrigation Method
+Geographic Region
+Academic Subject
+Material
+Product Brand
+Client Engagement Channel
+Transportation Infrastructure Type
+Beer Style
+Physical Activity
+Gemstone Type
+Pet Category
+Country
+Clothing Type
+Professional Sports Team
+Dance Discipline
+Musical Instruments
+Airline Name
+Building Type
+Household Appliance
+Award Medal Type
+Incident Type
+New York City Borough
+Occupation
+Cuisine
+Season (Calendar & Climatic)
+NCAA Division I Athletic Conference
+Demographic Group
+Media Asset Type
+Stage Musical Title
+Company Name
+Live Performance Type
+Art Period
+Animal Species
+Travel Destinations
+Event Type
+Mobile Phone Type
+Accommodation Type
+Operating System
+Internet, Cloud, and Digital TV Service Providers
+Vehicle Powertrain Type
+Streaming Service
+Superhero
+Weather Type
+Cooking Technique
+Religious Affiliation
+Sentiment
+Warehouse Storage Type
+Household Life Stage
+Football Clubs (Association Football)
+Biome Type
+Family Structure
+Playground Feature
+Handicraft Category
+Disease Name
+Retail Location Type
+Central Bank Name
+Semiconductor Manufacturing Stage
+Art and Craft Techniques
+Park Type
+Notable Creative Artists and Musical Acts
+Roadway Classification
+Nuclear Reactor Type
+Bird Species
+Camera Form Factor
+Island Name
+Residential Room Type
+Business Segment (Food & Beverage)
+Leisure Activity
+Museum Object Type
+Country / Sovereign State
+Pollution Source or Pollutant Type
+Medical Injury Type
+Dietary Preference or Restriction
+Dietary Protein Source
+Park Name (Theme & Amusement Parks and Major Public Parks)
+Coffee Brewing Method
+Social Event Type
+Purchase Channel
+Family Size (number of members)
+Investment Asset Class
+Expense Category
+Music Genre
+Dwelling Type
+Video Game Title
+Ecosystem Type
+Learning Delivery Mode
+Vehicle Model
+City
+Time of Day
+Produce (Fruit & Vegetable)
+Medical Procedure
+Major River Name
+Lighting Type
+Family Generational Role
+Landmark Name
+Sports Position
+Seafood Species
+Occupational Role
+Travel Category
+Marine Protected Area Name
+Color
+Protected Area Name
+Cryptocurrency Name
+Food Category
+Launch Outcome
+Jewelry Design Style
+Fashion Accessory Category
+Galaxy Type (morphology & class)
+Tourism Type
+Art Movement
+Currency Code
+U.S. National Park Name
+Theatrical Show Title
+Interest Category
+Financial Incentive Type
+Affectionate Gestures
+Cryptocurrency Exchange
+Mobile OS
+Lake Name
+Medical Specialty
+Plant Establishment Method
+Health Topic
+Higher Education Institution Type
+Eating Occasion
+Storage Type
+Zoning District (Land Use)
+Coffee Beverage / Preparation Style
+U.S. Battleground (Swing) States
+Environmental Impact Categories
+Education Program or Level
+Data Collection Method
+Game Genre
+Holiday Name
+Museum Focus
+Waste Material Type
+Agricultural Practice
+Freight Cargo Type
+Game Title
+Funding Source
+Customer Acquisition Channel
+Supply Chain Stage
+Civil Engineering Structure Type
+Recipe Ingredient
+Software Application Name
+Sofa Type
+Sales Offer Type
+Financial Product Type
+Festival Theme
+Camera Lens Type
+Concession Stand Product
+Climate Classification
+Primate Species
+Sports Facility Type
+Finishing Position
+Sustainability Category
+Innovation Focus Area
+Audio Listening Method
+Travel Market Segment
+Fitness Goal
+Communication Channel
+Social Determinants of Health
+Light Pollution Severity
+Tourist Attraction
+Grain Type
+Medical Condition
+Furniture Category
+Heritage Site Type
+Beverage Type
+Photography Genre
+Louisiana Parish
+North American Bird Flyways
+Research Publication Type
+Education Level
+Earring Type
+Common Illicit and Recreational Drugs
+Child Developmental Skill Category
+Occasion (Usage Scenario)
+Biomass Feedstock Type
+Major Animal Groups
+Alcoholic Beverage Type
+Establishment Type
+Interior Design Style
+Website Content Section
+Property Flooring Type
+Health Condition Category
+Membership Type
+Vehicle Connected Services
+Satellite Mission Type
+Vehicle Propulsion Type
+Marketing Channel
+Outreach Channel
+Retail Store Category
+Building Siding Material
+Fantasy Creature or Race
+Data Source Type
+Milestone Type
+Cyberattack Type
+Art Subject
+Citrus Variety
+Food Item
+Systems of Medicine
+Insect Common Name
+Major Sporting Events and Championships
+Insurance Type
+Major Entertainment Award Ceremonies
+Price Level
+Environmental Mitigation Methods
+Venue Seating Section
+Home Decor Category
+Cancer Type
+Driving Environment
+Wearable Accessory Type
+Play Activity Type
+Reef Site
+Music Distribution Format
+Sports Court Surface
+Application Category (App Store)
+Blockchain Platform
+Exercise Type
+Streaming Platforms
+Gemstone Color
+Cultivation Environment
+Sensor Type
+Wine Grape Varieties and Styles
+Sport Competition Tier
+Email Service Provider
+Costume Character
+Livestock Species
+Vineyard/Appellation (Wine Region)
+Historical Period
+Class Type
+Permit Sector (Construction/Development)
+Religious Symbol
+Level of Care (Senior / Long-Term Care)
+Cost Element (Accounting / Project Cost Category)
+Fruit Name
+Pollution Source
+Dietary Food Group
+Major Agricultural Regions (global)
+Neighborhood Name
+Manufacturing Step
+Amusement Ride Type
+Gaming Publication
+Dish Type
+Retailer Name
+Social & Environmental Impact Area
+Pesticide Type (Target/Function)
+Vehicle Maintenance Type
+Economic Region (industrial and economic clusters)
+Painting Medium
+Environmental Remediation Method
+Professional Skill
+Marketing Campaign Type
+Insulation Material
+Craft Supplies
+Geographic Region (Global, Continental, and Subregional)
+Major Global Shipping Passages and Chokepoints
+Project Phase
+Comic Book Issue
+Dominant Forest Cover Type
+Device Capabilities (Sensors, Connectivity & Features)
+Competition Gender Division
+Irrigation Source Type
+Gift Category
+Vessel Type
+Gallery Sector (Ownership / Institution Type)
+Notable Sacred Sites
+Online Video Platforms
+Agricultural Operation Type
+Tropical Cyclone Intensity Category (Saffir–Simpson and related classifications)
+Internet Connectivity Status
+Major Transportation Hub Cities
+Water Supply Type
+Performance Terrain Type
+Coral Reef Region
+Workout Modality
+Common Pest Type
+Patrol Method
+Library Name (Major Public, National, and Research Libraries)
+Customer Segment
+Sustainable Building Feature
+Subscription Action
+Fatal Incident Type
+Consumer Segment
+Learning Resource Type
+Service Category
+Manufacturing Process
+Pest Control Technique
+Personal Financial Concern Category
+Software Application
+Result Status (Test/Operation)
+Agricultural Support Category
+Weightlifting and Strength Training Exercises
+Access Level
+Therapeutic Area
+Employment Type
+Event Category
+Wellness Program Type
+Packaging Material
+Major Disaster Events
+ADAS Feature
+Military Platform Type
+Subscription Plan Tier
+Public Policy Areas
+Organization Sector
+Notable National Leaders (Heads of State or Government)
+Planetary Rover Name
+Prefectures of Japan
+Ballet Title
+Project Category
+Space Mission
+Footwear Style
+Commercial Aircraft Model
+Environmental Management Strategies
+Forage Type
+Clinical Service Type
+Tree Pruning System
+Hat Style
+Zoning/Management Zone Type
+U.S. Military Service Branch
+App Permission
+News Organization
+Ranching Practices and Systems
+Holding Institution or Collection
+Construction & Infrastructure Project Type
+Landscape Corridor or Long-Distance Trail Name
+Home Feature
+Production Method
+Pricing Basis
+Booking Channel
+Educational Focus Area
+Ecosystem / Habitat Type
+Craft Materials
+Subsystem / Component Type
+Flavor Profile (Tasting Descriptors)
+Transport Route Type
+Conservation Organization
+Basic Human Needs
+Fashion Style
+Humanitarian Aid Sector
+Character Type
+Flag State (Country of Registration)
+Aircraft Category (type/market segment)
+Technology Solution Type
+Roofing Material/Type
+Major Fishery Locations
+Instructional Method
+Common Amphibian Species (common names)
+Mammal Species Name
+Insect Group (common names)
+Major U.S. Rivers
+Major League Baseball Ballpark
+Movie Theater Format
+Major Rail and Metro Stations
+Food Item Name
+Medical Treatment Type
+Environmental Management Technique
+Handbag Style
+Artwork Title
+Personal Protective Equipment (PPE) Type
+Academic Subject Category
+Microphone Type
+Space Agency
+Pollinator Types
+Countries Currently Affected by Armed Conflict or Political Violence
+Computer Peripherals
+U.S. Federal Agency
+Agricultural Activities
+Government Budget Category
+Mission Phase
+Certification Level
+Leisure Amenities
+Personal Relationship Type
+Astronomical Observatory Name
+National Capital Cities
+Cultural Institution Type
+Historic Site Name
+Medical Interventions
+Types of Environmental Regulations
+Loyalty Program Feature
+Racing Circuit Type
+Competitor Brand (Fashion & Apparel)
+Hazard Type
+Voter Segment
+Pet Service Type
+Pet Wellness Package Type
+Benefit Recipient Type
+Forms of Folklore
+Generational Cohort
+Vehicle Part Type
+Ad Format
+Jewelry-making Technique
+Theme Park Name
+Cultural Attire
+Chinese New Year Foods
+Maize (Corn) Cultivar/Variety
+Software Functional Area
+U.S. Historic Site Name
+Consumer Electronics & Computer Hardware Category
+Educational Program Type
+Art Installation Type
+Culinary Trend
+Tour Activity
+International Trade Barrier Type
+Server Role
+Emotion Type
+Cause of Damage
+Passenger Persona (Travel Behavior)
+Solar System Planet
+Healthcare Facility Type
+Application Functionality
+Role-Playing Game System
+Freight Transport Mode
+Plantation Companies (Palm Oil & Tropical Crops)
+Product Variant (Production Method)
+Subsidy Program Name
+Accommodation Property Name
+Personal Data Type
+Financial Transaction Type
+Spending Occasion
+Home Improvement Project Type
+Consumer Market Segment
+Cultural Art Traditions
+Dimensions of Wellbeing
+Ticket Category
+Hardware Component Type
+Manufacturing Machine Type
+Motorcycle Type
+Light Color (illumination)
+Named Coasts and Coastal Regions (global)
+Endorsement Category (Products & Services)
+Communication Purpose
+Motor Vehicle Collision Type
+Cultural Offering Type
+Donor Classification
+Lemur Taxon Name (Genus and Species)
+Therapy Modality (Physical, Psychological, and Complementary Therapies)
+Body Shape (Figure Type)
+Comic Book Series
+Outdoor Lighting Fixture Type
+Material Sourcing Type
+Power Plant
+Driving Scenario
+Flight Route (region-to-region categories)
+Skilled Trade
+Magazine Name
+Land Use / Development Type
+Vehicle Model and Trim
+Ritual Type
+High-Risk Patient Groups
+Consumer & Lifestyle Trends
+TV Network (U.S. major broadcast & cable)
+Regulatory Policy Area
+Subject Area (Topic)
+Exhibit Theme
+Venue Type
+Toy Type
+Proximity Level
+Harvesting Method
+Smartphone Model
+Seabird Species
+Hazard Mitigation Project Type
+Accessibility Feature Type
+Tractor Powertrain & Control Configuration
+Major Orchard Regions (Apple & Tree-Fruit Producing Regions)
+Filtration Technology
+Running Shoe Segment
+Art & Craft Workshop Discipline
+Scientific Discovery Category
+Monetization Method
+Area of Specialization
+Company Size Category (by number of employees)
+Debris Material
+Fishing Capture Method
+Land Management Practice
+Basketball Shot Zone
+Dress Silhouette
+Water Conservation Measure
+Highest Educational Attainment
+Love Language Type
+Crop Pest
+Gift Type
+Farm Diversification Strategies
+Payment Method
+Climate Zone (major types — Köppen & common names)
+Architectural Style
+Schedule Disruption Reason
+Tomato Variety
+Lighting Fixture Type
+Washing Machine Type
+Art Supply Set Type
+Gender Identity
+Marital Status

icon_generation/backup/enhance_domains.py

@@ -0,0 +1,551 @@
+#!/usr/bin/env python3
+"""
+处理 domain_value_pairs_filtered.txt
+1. 按 domain 聚合
+2. 只保留 accepted_domains.txt 中的 domains
+3. 使用 LLM 补全 specific values 并按 real-world frequency 排序
+4. 丢弃过于冷门、罕见的 values
+"""
+
+import json
+import os
+import requests
+from typing import List, Dict, Tuple, Optional
+from collections import defaultdict
+import time
+import threading
+from concurrent.futures import ThreadPoolExecutor, as_completed
+
+
+class ValueEnhancer:
+    """使用 LLM 增强和排序 values"""
+    
+    def __init__(self, api_key=None, base_url=None, model=None):
+        """初始化 LLM analyzer"""
+        self.api_key = api_key or os.getenv("OPENAI_API_KEY") or os.getenv("AIHUBMIX_API_KEY", "")
+        self.base_url = base_url or os.getenv("OPENAI_BASE_URL", "https://aihubmix.com/v1")
+        self.model = model or os.getenv("OPENAI_MODEL", "gpt-5-mini")
+        self.lock = threading.Lock()  # 线程锁，用于打印
+        self.completed_count = 0  # 已完成计数
+        self.total_domains = 0  # 总 domain 数
+    
+    def enhance_and_sort_values(self, domain: str, values: List[Tuple[str, int]], idx: int, total: int, start_time: float) -> Dict:
+        """
+        使用 LLM 增强和排序 values
+        
+        Args:
+            domain: domain 名称
+            values: [(value, count), ...] 列表
+            idx: 当前索引
+            total: 总数
+            start_time: 开始时间
+        
+        Returns:
+            {
+                'enhanced_values': [(value, estimated_frequency), ...],
+                'reasoning': 'explanation',
+                'domain': domain
+            }
+        """
+        # 显示开始处理
+        with self.lock:
+            progress = (idx / total) * 100
+            elapsed = time.time() - start_time
+            avg_time = elapsed / idx if idx > 0 else 0
+            remaining = avg_time * (total - idx)
+            
+            print(f"\n{'=' * 80}")
+            print(f"🔄 处理 {idx}/{total} ({progress:.1f}%): {domain}")
+            print(f"⏱️  已用时间: {elapsed:.1f}秒 | 预计剩余: {remaining:.1f}秒")
+            print(f"   原始 values: {len(values)} 个")
+            print(f"   原始总 count: {sum(c for _, c in values):,}")
+            
+            # 显示前 3 个示例
+            print(f"   示例值:")
+            for i, (value, count) in enumerate(sorted(values, key=lambda x: x[1], reverse=True)[:3]):
+                print(f"      {i+1}. {value} ({count:,})")
+            
+            print(f"   🤖 调用 LLM 进行增强和排序...")
+        
+        prompt = self._build_enhancement_prompt(domain, values)
+        
+        try:
+            response = self._query_llm(prompt)
+            
+            if response:
+                # 清理可能的 markdown 代码块
+                cleaned_response = response.strip()
+                if cleaned_response.startswith('```'):
+                    lines = cleaned_response.split('\n')
+                    cleaned_response = '\n'.join(lines[1:-1] if lines[-1].strip() == '```' else lines[1:])
+                    cleaned_response = cleaned_response.replace('```json', '').replace('```', '').strip()
+                
+                result = json.loads(cleaned_response)
+                
+                # 验证返回格式
+                if 'enhanced_values' in result:
+                    enhanced_values = result['enhanced_values']
+                    reasoning = result.get('reasoning', '')
+                    refined_domain_name = result.get('refined_domain_name')
+                    
+                    # 如果有新的 domain name，使用新的
+                    final_domain = refined_domain_name if refined_domain_name else domain
+                    
+                    with self.lock:
+                        self.completed_count += 1
+                        progress = (self.completed_count / total) * 100
+                        
+                        # 显示 domain name 变化（如果有）
+                        if refined_domain_name and refined_domain_name != domain:
+                            print(f"   🔄 Domain 名称优化:")
+                            print(f"      原始: {domain}")
+                            print(f"      优化: {refined_domain_name}")
+                        
+                        print(f"   ✅ 增强完成!")
+                        print(f"   增强后 values: {len(enhanced_values)} 个")
+                        print(f"   变化: {len(enhanced_values) - len(values):+d}")
+                        print(f"   理由: {reasoning}")
+                        
+                        # 显示前 3 个增强后的值
+                        print(f"   增强后 Top 3:")
+                        for i, item in enumerate(enhanced_values[:3], 1):
+                            value = item['value']
+                            freq = item['estimated_frequency']
+                            print(f"      {i}. {value} (frequency: {freq:,})")
+                        
+                        print(f"   总进度: {self.completed_count}/{total} ({progress:.1f}%)")
+                    
+                    return {
+                        'enhanced_values': enhanced_values,
+                        'reasoning': reasoning,
+                        'domain': final_domain,  # 使用优化后的 domain name
+                        'original_domain': domain  # 保留原始 domain name
+                    }
+                else:
+                    with self.lock:
+                        self.completed_count += 1
+                        print(f"      ⚠️  LLM 响应缺少必需字段")
+                    # 返回原始数据（格式统一）
+                    return {
+                        'enhanced_values': [
+                            {'value': v, 'estimated_frequency': c}
+                            for v, c in values
+                        ],
+                        'reasoning': 'LLM response format error, kept original',
+                        'domain': domain
+                    }
+                    
+            else:
+                with self.lock:
+                    self.completed_count += 1
+                    print(f"      ⚠️  LLM API 调用失败")
+                return {
+                    'enhanced_values': [
+                        {'value': v, 'estimated_frequency': c}
+                        for v, c in values
+                    ],
+                    'reasoning': 'LLM API failed, kept original',
+                    'domain': domain
+                }
+                
+        except json.JSONDecodeError as e:
+            with self.lock:
+                self.completed_count += 1
+                print(f"      ⚠️  LLM 响应不是有效的 JSON: {e}")
+                print(f"      响应: {response[:300]}...")
+            return {
+                'enhanced_values': [
+                    {'value': v, 'estimated_frequency': c}
+                    for v, c in values
+                ],
+                'reasoning': 'JSON decode error, kept original',
+                'domain': domain
+            }
+        except Exception as e:
+            with self.lock:
+                self.completed_count += 1
+                print(f"      ⚠️  增强错误: {e}")
+            return {
+                'enhanced_values': [
+                    {'value': v, 'estimated_frequency': c}
+                    for v, c in values
+                ],
+                'reasoning': f'Error: {str(e)}, kept original',
+                'domain': domain
+            }
+    
+    def _build_enhancement_prompt(self, domain: str, values: List[Tuple[str, int]]) -> str:
+        """构建增强的 prompt"""
+        
+        # 构建现有 values 列表
+        values_str = '\n'.join([
+            f"  - {value} (current count: {count})"
+            for value, count in values[:50]  # 最多显示前50个
+        ])
+        
+        if len(values) > 50:
+            values_str += f"\n  ... and {len(values) - 50} more values"
+        
+        prompt = f"""
+You are a data curation and domain expert. Given a domain and its current list of values, your task is to produce a clean, comprehensive, and realistic set of values by following the steps below.
+
+Your responsibilities:
+
+0. **REVIEW DOMAIN NAME**:
+   - First, evaluate whether the domain name accurately describes the values.
+   - If the domain name is ambiguous, too broad, too narrow, or doesn't properly represent the values, suggest a better, more accurate domain name.
+   - The refined domain name should clearly indicate what category/dimension these values represent.
+   - Example: "Category" → "Product Category"; "Type" → "Vehicle Type"; "Name" → "Brand Name"
+
+1. **KEEP** all existing values that are meaningful, commonly known, and relevant to the domain.
+2. **ADD** important missing values that are widely recognized in the real world for this domain.
+3. **REMOVE** values that are obscure, extremely niche, rarely used, outdated, or not commonly recognized.
+4. **DEDUPLICATE** values:
+   - Remove exact duplicates.
+   - Merge near-duplicates or synonyms (e.g., spelling variants, abbreviations, singular/plural forms).
+   - Normalize values so that each real-world concept appears **only once**.
+5. **REMOVE REDUNDANT ATTRIBUTES**:
+   - If values contain extra attributes (e.g., qualifiers, parenthetical notes, versions, descriptors),
+     keep only the **canonical, clean value name** unless the attribute is essential to distinguish meaning.
+   - Example: "Football (Soccer)" → "Soccer"; "iPhone 14 Pro Max" → "iPhone 14 Pro" (if variants are not required).
+6. **ESTIMATE** a real-world frequency/popularity score for each final value (scale: 1–10000),
+   reflecting how commonly the value is encountered or recognized in practice.
+7. **SORT** all final values by estimated real-world frequency, from most common/popular to least.
+
+Domain:
+"{domain}"
+
+Current Values:
+{values_str}
+
+Guidelines:
+- For well-defined enumerable domains (e.g., US States, Countries), include **all standard items**.
+- For category-style domains (e.g., Sports, Genres, Product Types), focus on **mainstream, widely recognized** values.
+- Prefer canonical names over aliases or variants.
+- Base frequency estimates on real-world usage, awareness, or prevalence — not on the input list.
+- Add missing values that a knowledgeable human would reasonably expect to see.
+- Aim for **20–50 values** for most domains (more only when the domain is inherently exhaustive).
+
+Output Requirements:
+- Return your response in the following JSON format ONLY.
+- Do NOT include markdown, comments, or explanatory text outside the JSON.
+- Ensure the final list is fully deduplicated and normalized.
+- If the domain name is appropriate, set "refined_domain_name" to null or the same name.
+- If the domain name should be changed, provide a better, more specific name in "refined_domain_name".
+
+{{
+  "refined_domain_name": "Better Domain Name" or null,
+  "enhanced_values": [
+    {{"value": "ValueName1", "estimated_frequency": 10000}},
+    {{"value": "ValueName2", "estimated_frequency": 8500}}
+  ],
+  "reasoning": "Brief explanation of domain name refinement (if any), key additions, removals, deduplication decisions, and sorting rationale."
+}}
+
+Examples:
+
+Example — Domain: "US State"
+- DOMAIN NAME: Appropriate, keep as is (refined_domain_name: null)
+- KEEP: All valid US states already present
+- ADD: Any missing states to complete the full set of 50
+- REMOVE: None
+- DEDUPLICATE: Merge variants like "CA" and "California" → "California"
+- SORT: By population (California, Texas, Florida, New York...)
+
+Example — Domain: "Sport Type"
+- DOMAIN NAME: Appropriate, keep as is (refined_domain_name: null)
+- KEEP: Soccer, Basketball, Baseball, Tennis
+- ADD: Football, Cricket, Swimming (if missing and globally popular)
+- REMOVE: Extremely obscure sports unless they are regionally mainstream
+- DEDUPLICATE: Merge "Football" and "American Football" appropriately based on context
+- SORT: By global popularity (Soccer, Basketball, Cricket, Tennis...)
+
+Example — Domain: "Category" (with values like "Electronics", "Books", "Clothing")
+- DOMAIN NAME: Too generic → refined_domain_name: "Product Category"
+- Values processing continues as normal...
+
+Example — Domain: "Type" (with values like "SUV", "Sedan", "Truck")
+- DOMAIN NAME: Too generic → refined_domain_name: "Vehicle Type"
+- Values processing continues as normal...
+"""
+        
+        return prompt
+    
+    def _query_llm(self, prompt: str) -> Optional[str]:
+        """查询 LLM API"""
+        headers = {
+            'Authorization': f'Bearer {self.api_key}',
+            'Content-Type': 'application/json'
+        }
+        
+        data = {
+            'model': self.model,
+            'messages': [
+                {
+                    'role': 'system', 
+                    'content': 'You are a data expert specialized in enhancing and organizing domain-specific values based on real-world knowledge. Always return valid JSON format only, without any markdown formatting or extra text.'
+                },
+                {
+                    'role': 'user', 
+                    'content': prompt
+                }
+            ],
+            'temperature': 0.3
+        }
+        
+        try:
+            response = requests.post(
+                f'{self.base_url}/chat/completions',
+                headers=headers,
+                json=data,
+                timeout=120
+            )
+            response.raise_for_status()
+            
+            result = response.json()
+            return result['choices'][0]['message']['content'].strip()
+            
+        except requests.exceptions.Timeout:
+            print("      ❌ LLM API 超时")
+            return None
+        except requests.exceptions.HTTPError as e:
+            print(f"      ❌ LLM API HTTP 错误: {e}")
+            return None
+        except requests.exceptions.RequestException as e:
+            print(f"      ❌ LLM API 请求错误: {e}")
+            return None
+        except KeyError as e:
+            print(f"      ❌ LLM API 响应格式错误: {e}")
+            return None
+
+
+def parse_csv_line(line: str) -> Tuple[str, str, int]:
+    """解析 CSV 行"""
+    parts = []
+    current = []
+    in_quotes = False
+    
+    for char in line:
+        if char == '"':
+            in_quotes = not in_quotes
+        elif char == ',' and not in_quotes:
+            parts.append(''.join(current).strip())
+            current = []
+        else:
+            current.append(char)
+    parts.append(''.join(current).strip())
+    
+    if len(parts) >= 3:
+        domain = parts[0]
+        attribute = parts[1]
+        count = int(parts[2])
+        return domain, attribute, count
+    else:
+        return None, None, 0
+
+
+def load_accepted_domains(file_path: str) -> List[str]:
+    """加载可接受的 domains"""
+    domains = []
+    with open(file_path, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                continue
+            domains.append(line)
+    return domains
+
+
+def aggregate_by_domain(file_path: str) -> Dict[str, List[Tuple[str, int]]]:
+    """按 domain 聚合数据"""
+    domain_data = defaultdict(list)
+    
+    with open(file_path, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                continue
+            
+            domain, attribute, count = parse_csv_line(line)
+            if domain:
+                domain_data[domain].append((attribute, count))
+    
+    return domain_data
+
+
+def main():
+    print("=" * 80)
+    print("Domain Values 增强和排序工具")
+    print("=" * 80)
+    print()
+    
+    # 1. 加载可接受的 domains
+    print("📖 读取可接受的 domains...")
+    accepted_domains = load_accepted_domains('accepted_domains.txt')
+    print(f"✅ 可接受的 domains: {len(accepted_domains)} 个")
+    if len(accepted_domains) <= 20:
+        for idx, domain in enumerate(accepted_domains, 1):
+            print(f"   {idx}. {domain}")
+    else:
+        print(f"   前 10 个: {', '.join(accepted_domains[:10])}")
+        print(f"   ... 还有 {len(accepted_domains) - 10} 个")
+    print()
+    
+    # 2. 加载和聚合数据
+    print("📖 读取和聚合数据...")
+    domain_data = aggregate_by_domain('domain_value_pairs_filtered.txt')
+    print(f"✅ 总共 {len(domain_data)} 个不同的 domains")
+    print()
+    
+    # 3. 过滤只保留接受的 domains
+    filtered_data = {
+        domain: values 
+        for domain, values in domain_data.items() 
+        if domain in accepted_domains
+    }
+    
+    print(f"🔍 过滤后保留 {len(filtered_data)} 个 domains")
+    if len(filtered_data) <= 20:
+        for domain in filtered_data.keys():
+            value_count = len(filtered_data[domain])
+            total_count = sum(c for _, c in filtered_data[domain])
+            print(f"   - {domain}: {value_count} values, total count: {total_count:,}")
+    else:
+        print("   前 10 个 domains:")
+        for i, domain in enumerate(list(filtered_data.keys())[:10], 1):
+            value_count = len(filtered_data[domain])
+            total_count = sum(c for _, c in filtered_data[domain])
+            print(f"      {i}. {domain}: {value_count} values, total count: {total_count:,}")
+        print(f"   ... 还有 {len(filtered_data) - 10} 个 domains")
+    print()
+    
+    # 4. 初始化增强器
+    enhancer = ValueEnhancer()
+    enhancer.total_domains = len(filtered_data)
+    
+    # 5. 准备任务列表
+    print("=" * 80)
+    print("🚀 开始处理和增强 domains (10线程并发)")
+    print("=" * 80)
+    print()
+    
+    # 转换为列表以便索引
+    domain_items = list(filtered_data.items())
+    total_domains = len(domain_items)
+    
+    enhanced_results = {}
+    start_time = time.time()
+    
+    # 6. 使用线程池并发处理
+    num_threads = 15
+    print(f"🔧 使用 {num_threads} 个线程并发处理")
+    print()
+    
+    with ThreadPoolExecutor(max_workers=num_threads) as executor:
+        # 提交所有任务
+        future_to_domain = {
+            executor.submit(
+                enhancer.enhance_and_sort_values,
+                domain,
+                values,
+                idx,
+                total_domains,
+                start_time
+            ): domain
+            for idx, (domain, values) in enumerate(domain_items, 1)
+        }
+        
+        # 收集结果
+        domain_name_mapping = {}  # 原始 domain -> 优化后 domain 的映射
+        
+        for future in as_completed(future_to_domain):
+            original_domain = future_to_domain[future]
+            try:
+                result = future.result()
+                final_domain = result['domain']
+                enhanced_results[final_domain] = result['enhanced_values']
+                
+                # 记录 domain name 映射
+                if final_domain != original_domain:
+                    domain_name_mapping[original_domain] = final_domain
+                    
+            except Exception as e:
+                print(f"❌ 处理 {original_domain} 时出错: {e}")
+                # 保留原始数据
+                enhanced_results[original_domain] = [
+                    {'value': v, 'estimated_frequency': c}
+                    for v, c in filtered_data[original_domain]
+                ]
+    
+    # 6. 保存结果
+    elapsed_time = time.time() - start_time
+    
+    print()
+    print("=" * 80)
+    print("💾 保存结果...")
+    print("=" * 80)
+    
+    # 显示 domain name 变化（如果有）
+    if domain_name_mapping:
+        print()
+        print(f"🔄 Domain 名称优化记录 ({len(domain_name_mapping)} 个):")
+        for original, refined in domain_name_mapping.items():
+            print(f"   {original} → {refined}")
+        print()
+    
+    output_file = 'domain_value_pairs_enhanced.txt'
+    
+    # 展开所有 domain-value pairs 并按 frequency 全局排序
+    all_pairs = []
+    for domain, values in enhanced_results.items():
+        for item in values:
+            value = item['value']
+            freq = item['estimated_frequency']
+            all_pairs.append((domain, value, freq))
+    
+    # 按 frequency 降序排序
+    all_pairs.sort(key=lambda x: x[2], reverse=True)
+    
+    # 保存到文件
+    with open(output_file, 'w', encoding='utf-8') as f:
+        for domain, value, freq in all_pairs:
+            # 处理可能包含逗号的字段
+            if ',' in value:
+                value = f'"{value}"'
+            if ',' in domain:
+                domain = f'"{domain}"'
+            f.write(f"{domain},{value},{freq}\n")
+    
+    print(f"✅ 结果已保存到: {output_file}")
+    print()
+    
+    # 7. 统计信息
+    print("=" * 80)
+    print("📊 统计信息:")
+    print("=" * 80)
+    print(f"处理的 domains: {len(enhanced_results)}")
+    print(f"总 pairs 数: {len(all_pairs)}")
+    print(f"总耗时: {elapsed_time:.1f} 秒")
+    print()
+    
+    # 按 domain 显示统计
+    print("各 Domain 统计:")
+    for domain in accepted_domains:
+        if domain in enhanced_results:
+            count = len(enhanced_results[domain])
+            print(f"   {domain}: {count} values")
+    print()
+    
+    # 显示 Top 20 pairs
+    print("🏆 Top 20 Pairs (按 estimated frequency):")
+    for i, (domain, value, freq) in enumerate(all_pairs[:20], 1):
+        print(f"   {i:2d}. {domain}: {value} (frequency: {freq:,})")
+    
+    print()
+    print("=" * 80)
+    print("✨ 增强完成！")
+    print("=" * 80)
+
+
+if __name__ == '__main__':
+    main()

icon_generation/backup/enhance_log.txt

@@ -0,0 +1,1130 @@
+================================================================================
+Domain Values 增强和排序工具
+================================================================================
+
+📖 读取可接受的 domains...
+✅ 可接受的 domains: 609 个
+   1. US State
+   2. Entertainment Genre
+   3. Sport Type
+   4. Digital Platform
+   5. Industry Sector
+   6. Gender Demographic
+   7. Product Category
+   8. Canadian Province
+   9. Crop Type
+   10. Geographic Classification
+   11. Sports Team
+   12. Animal Species
+   13. Browser Type
+   14. Device Type
+   15. School Category
+   16. Art Medium
+   17. Material
+   18. Specialization Area
+   19. Energy Technology
+   20. Trade Direction
+   21. Dog Breed
+   22. Continent
+   23. Energy Source
+   24. Occurrence Frequency
+   25. Economic System
+   26. Country
+   27. Transportation Mode
+   28. Travel Location
+   29. Company Name
+   30. Income Bracket
+   31. Sports League
+   32. Smart Home Category
+   33. Media Type
+   34. Academic Subject
+   35. Land Use Type
+   36. Vehicle Classification
+   37. Irrigation Method
+   38. Calendar Month
+   39. Socioeconomic Status Level
+   40. Farm Size Category
+   41. Digital Service Provider
+   42. Political Party
+   43. TV Network
+   44. Physical Activity
+   45. Dance Discipline
+   46. Musical Instrument
+   47. Gemstone Type
+   48. Building Classification
+   49. Product Brand
+   50. Monetization Method
+   51. Airline Name
+   52. Clothing Type
+   53. Football Club
+   54. Cuisine
+   55. Beer Style
+   56. Diplomatic Mission Type
+   57. Political Alignment
+   58. Airport Code
+   59. Operation Type
+   60. Home Appliance
+   61. Seaport
+   62. Client Engagement Channel
+   63. Attraction Park Name
+   64. Farming System
+   65. Superhero
+   66. Disease Name
+   67. Occupational Role
+   68. Organization Type
+   69. Infrastructure Type
+   70. Athletic Conference
+   71. Integrity Standing
+   72. Toy Type
+   73. Creative Artist
+   74. Administrative Borough
+   75. Biome Type
+   76. Event Classification
+   77. Seafood Species
+   78. Tea Type
+   79. Leisure Activity
+   80. Expense Type
+   81. Smartphone Brand
+   82. Landmark Name
+   83. Audience Segment
+   84. Occupation
+   85. Calendar Season
+   86. Subject Area
+   87. Incident Type
+   88. Tillage System
+   89. Protected Area Name
+   90. Art Technique
+   91. Fabric Type
+   92. Extreme Temperature Type
+   93. Island Name
+   94. Podcast Genre
+   95. Time Off Type
+   96. Pet Category
+   97. Medal Type
+   98. Opinion Type
+   99. Playground Feature
+   100. Video Game Title
+   101. Regulatory Domain
+   102. Streaming Service
+   103. Vehicle Model
+   104. Game Name
+   105. Demographic Group
+   106. Cooking Technique
+   107. Media Asset Type
+   108. Bird Species
+   109. Religious Affiliation
+   110. Exhibit Theme
+   111. Medical Specialty
+   112. Weather Type
+   113. Artifact Class
+   114. Brewing Method
+   115. Dwelling Type
+   116. Company Size Category
+   117. City
+   118. Color
+   119. Family Structure Type
+   120. Performance Type
+   121. Craft Category
+   122. Nutritional Component
+   123. Music Genre
+   124. Venue Type
+   125. Marine Conservation Name
+   126. Pollution Cause Type
+   127. Medical Procedure
+   128. Property Type
+   129. Radio Genre
+   130. Art Period
+   131. Phone Model
+   132. Business Segment
+   133. Soil Texture
+   134. Central Bank Name
+   135. Cover Crop Type
+   136. Road Classification
+   137. Dietary Consideration
+   138. Beverage Type
+   139. Vehicle Powertrain
+   140. Lake Name
+   141. River Basin Name
+   142. Accessory Category
+   143. Jewelry Design Style
+   144. Agricultural Practice
+   145. Ecosystem Type
+   146. Home Space Type
+   147. Marital Status
+   148. Building Style
+   149. Day Classification
+   150. Produce Name
+   151. Medical Injury Type
+   152. Risk Rating
+   153. Payment Method
+   154. Education Category
+   155. Proficiency Level
+   156. Investment Asset Class
+   157. Park Category
+   158. Purchase Channel
+   159. Protein Type
+   160. Class Classification
+   161. Casualty Status
+   162. Phone Type
+   163. Camera Form Factor
+   164. Travel Category
+   165. Recipe Ingredient
+   166. Flight Disruption Type
+   167. Musical Title
+   168. Food Category
+   169. Generational Role
+   170. National Park Name
+   171. Show Title
+   172. Interest Category
+   173. Supply Chain Stage
+   174. Comedy Type
+   175. Family Member Count
+   176. Metal Type
+   177. Historical Period
+   178. Day Period
+   179. Manufacturing Stage
+   180. Retail Environment
+   181. Crypto Exchange
+   182. Customer Segment Name
+   183. Holiday Name
+   184. Recipient Category
+   185. Minority Indicator
+   186. Medical Condition
+   187. Louisiana Parish
+   188. Performance Indicator
+   189. Sports Facility Type
+   190. Pest Category
+   191. Propulsion Type
+   192. Nuclear Reactor Type
+   193. Game Genre
+   194. Revenue Type
+   195. Financial Product Type
+   196. Galaxy Type
+   197. Environmental Impact
+   198. Craft Supply
+   199. Temperature Record Type
+   200. Furniture Category
+   201. Fruit Name
+   202. Social Event Type
+   203. Neighborhood
+   204. Service Category
+   205. Zoning District
+   206. Play Type
+   207. Wine Classification
+   208. Eating Occasion
+   209. Operating System
+   210. Career Level
+   211. Music Subgenre
+   212. Aircraft Model
+   213. Major Sports Event
+   214. Disaster Event
+   215. Harvest Method
+   216. Marine Area Name
+   217. Involvement Level
+   218. Gemstone Color
+   219. Tourist Attraction
+   220. Fantasy Entity
+   221. Household Life Stage
+   222. Climate Type
+   223. Air Quality Parameter
+   224. News Organization
+   225. Heritage Site Type
+   226. Earring Type
+   227. Fertilizer Type
+   228. Health Area
+   229. Conservation Status
+   230. Education Format
+   231. System Component
+   232. Gaming Publication
+   233. Animal Type
+   234. Animal Adoption Source
+   235. Capture Method
+   236. Festival Theme
+   237. Participation Level
+   238. Climate Zone Type
+   239. Achievement Award Category
+   240. Capital City
+   241. Waste Material
+   242. Outreach Channel
+   243. Reef Site
+   244. Exercise Type
+   245. Cancer Type
+   246. Content Streaming Platform
+   247. Turf Type
+   248. Home Decor Category
+   249. Geographic Region
+   250. Establishment Type
+   251. Mitigation Method
+   252. Grant Active
+   253. Storage Category
+   254. Concession Product
+   255. Citrus Variety
+   256. Coffee Style
+   257. Feedstock Type
+   258. Insurance Type
+   259. Land Management Practice
+   260. Application Category
+   261. Lighting Fixture Type
+   262. Data Domain
+   263. Award Ceremony
+   264. Conservation Organization
+   265. Music Distribution Format
+   266. Social Media Intensity
+   267. Engagement
+   268. Primate Species
+   269. Grain Category
+   270. Tourism Type
+   271. Launch Status
+   272. Funding Source
+   273. Email Provider
+   274. Amusement Ride Type
+   275. Vessel Classification
+   276. Fatal Incident Type
+   277. Mission Complexity Level
+   278. Mobile OS
+   279. Museum Focus
+   280. Market Trend
+   281. Enforcement Degree
+   282. Flyway Region
+   283. Travel Segment
+   284. Usage Scenario
+   285. Membership Category
+   286. Weightlifting Exercise
+   287. Lighting Type
+   288. Fitness Goal
+   289. Vineyard Designation
+   290. Food Product Name
+   291. Political Entity
+   292. Economic Region
+   293. Health Coverage Type
+   294. Data Collection Method
+   295. Innovation Area
+   296. Food Item
+   297. Drug Name
+   298. Store Category
+   299. Wearable Item Type
+   300. Photography Genre
+   301. Tomato Variety
+   302. Therapeutic Area
+   303. Sport Position
+   304. Packaging Material
+   305. Fashion Style
+   306. Cargo Type
+   307. Sensor Category
+   308. Public Policy
+   309. Military Platform Type
+   310. Plant Establishment Method
+   311. Marketing Channel
+   312. Education Level
+   313. Project Type
+   314. Listening Method
+   315. Social Determinant
+   316. Art Movement
+   317. Higher Education Type
+   318. Agricultural Type
+   319. Storage Type
+   320. Measure Type
+   321. ADAS Module
+   322. Debris Material
+   323. Retailer Name
+   324. Legislative Status
+   325. Engineering Structure Type
+   326. Pesticide Classification
+   327. Program Certification Level
+   328. Pandemic Period
+   329. Financial Incentive Type
+   330. Dish Type
+   331. Impact Area
+   332. Shipping Passage
+   333. Proximity Level
+   334. Sofa Type
+   335. Content Section
+   336. Corridor Name
+   337. Fixture Design
+   338. Immigration Status
+   339. Project Category
+   340. Painting Medium
+   341. Gift Category
+   342. Sanction Measure
+   343. Forest Cover Type
+   344. Software Application
+   345. Currency Code
+   346. Insulation Material
+   347. Aid Sector
+   348. Health Condition Category
+   349. Art Subject
+   350. Vehicle Maintenance Type
+   351. Tractor Configuration
+   352. Rover Name
+   353. Product Variant
+   354. Student Proficiency
+   355. Costume Character
+   356. Event Category
+   357. Seabird Species
+   358. Service Branch
+   359. Flag State Name
+   360. Insect Common Name
+   361. Accessibility Feature Type
+   362. Camera Lens Type
+   363. Roofing Type
+   364. Power Facility
+   365. Comic Issue
+   366. Sustainability Category
+   367. Court Surface
+   368. Consumer Profile
+   369. Technology Solution
+   370. Historic Site Name
+   371. Accommodation Name
+   372. Workout Modality
+   373. Season Phase
+   374. Satellite Function
+   375. Sales Offer Type
+   376. Driving Context
+   377. Professional Skill
+   378. GDP Component Type
+   379. App Pricing Model
+   380. Cultural Attire
+   381. Planet Name
+   382. Dietary Food Group
+   383. Technique Type
+   384. Wellbeing Dimension
+   385. Livestock Category
+   386. Care Level
+   387. Campaign Type
+   388. Shot Zone
+   389. Acquisition Channel
+   390. Work Gear Type
+   391. Light Pollution Severity
+   392. Source Collection
+   393. App Permission
+   394. Religious Symbol
+   395. Manufacturing Step
+   396. App Usage Level
+   397. Disruption Reason
+   398. Funding Type
+   399. Sustainable Feature
+   400. Green Space Accessibility
+   401. Fishery Location
+   402. Workshop Discipline
+   403. Developmental Skill Category
+   404. Cryptocurrency Name
+   405. Adoption Level
+   406. Hurricane Wind Category
+   407. Production Type
+   408. Pollution Source
+   409. Support Category
+   410. Amphibian Species Name
+   411. Sales Metric
+   412. Alcoholic Beverage Type
+   413. Property Flooring Type
+   414. Resource Classification
+   415. Orchard Region
+   416. Ballpark Name
+   417. Home Feature
+   418. Flight Route
+   419. Battleground State
+   420. Parental Involvement Level
+   421. Interaction Channel
+   422. Booking Point
+   423. Instructional Method
+   424. Space Agency
+   425. Polymer Material
+   426. Performance Standing
+   427. Connected Service
+   428. Cyber Attack Type
+   429. Irrigation Source Type
+   430. Organization Sector
+   431. Ballet Title
+   432. Hardware Category
+   433. CNY Food
+   434. Vehicle Part Type
+   435. Site Name
+   436. Price Level
+   437. Manufacturing Process
+   438. Participation Status
+   439. Affectionate Gesture
+   440. IP Infringement Type
+   441. Mitigation Project Type
+   442. Flavor Profile
+   443. Research Publication Type
+   444. Footwear Style
+   445. Environmental Strategy
+   446. Tributary Name
+   447. Cultivation Environment
+   448. Transit Station
+   449. Microphone Classification
+   450. Economic Indicator Type
+   451. Remediation Method
+   452. Project Phase
+   453. Financial Concern Category
+   454. Manure Management Method
+   455. Set Category
+   456. Education Focus
+   457. Permit Sector
+   458. Body Shape
+   459. Building Siding Material
+   460. Sport Tier
+   461. Filtration Technology
+   462. Theme Park Name
+   463. Manufacturing Machine Type
+   464. Space Mission
+   465. Agricultural Region
+   466. Essential Need
+   467. Craft Material
+   468. Management Technique
+   469. Car Model Name
+   470. Zone Designation
+   471. Design Style
+   472. Pruning Style
+   473. Sport Gender Division
+   474. Water Supply Type
+   475. Mammal Species Name
+   476. Insect Type
+   477. Product Form
+   478. Control Technique
+   479. Clinical Service
+   480. Component Type
+   481. Probability Basis
+   482. Hub City
+   483. Internet Access Status
+   484. Tariff Active
+   485. Hat Style
+   486. Pet Service Type
+   487. Computer Peripheral
+   488. Farming Diversification Strategy
+   489. Cost Element
+   490. Impact Severity
+   491. Washing Machine Form Factor
+   492. Spending Occasion
+   493. Video Platform
+   494. Election Type
+   495. Milestone Type
+   496. Index Basis
+   497. Loan Application Status
+   498. Farm Activity
+   499. Scenario Type
+   500. Gallery Sector
+   501. Forage Type
+   502. Ecological Area Type
+   503. Application Name
+   504. Patrol Method
+   505. Access Level
+   506. Result Status
+   507. Subsidy Indicator
+   508. Subscription Plan
+   509. Light Color
+   510. Wellness Program Type
+   511. Educational Attainment
+   512. Agricultural Pest
+   513. Library Name
+   514. Subject Category
+   515. Reef Region
+   516. Medicine System
+   517. National Leader
+   518. Observatory Name
+   519. Budget Category
+   520. Flood Risk Level
+   521. Pet Package Type
+   522. Voter Segment
+   523. Motorcycle Type
+   524. Message Purpose
+   525. Conflict Country
+   526. Cultural Institution Type
+   527. Mission Operation
+   528. Narrative Device
+   529. Japan Prefecture
+   530. Circuit Type
+   531. Cultural Offering
+   532. Character Kind
+   533. Medical Intervention
+   534. Hazard Classification
+   535. Shoe Segment
+   536. Software Function
+   537. Tour Activity
+   538. User Segment
+   539. Pricing Basis
+   540. Culinary Trend
+   541. Trade Barrier Type
+   542. Dress Silhouette
+   543. Performance Terrain
+   544. Aircraft Category
+   545. Competitor Brand
+   546. Discovery Category
+   547. Personal Data Type
+   548. Beneficiary Type
+   549. Program Type
+   550. Theater Format
+   551. Love Language Type
+   552. Pollinator Classification
+   553. Magazine Name
+   554. Patient Risk Group
+   555. Healthcare Facility Type
+   556. Endorsement Category
+   557. Subscription Action Type
+   558. Geographic Coast
+   559. Venue Seating Section
+   560. Worker Type
+   561. Route Type
+   562. Passenger Persona
+   563. Employment Model
+   564. Gift Type
+   565. Taxon Name
+   566. Therapy Modality
+   567. Driving Scenario
+   568. Blockchain Platform
+   569. Federal Agency
+   570. Environmental Regulation
+   571. Ranching Approach
+   572. Personal Relationship
+   573. Sourcing Material Type
+   574. Comic Series
+   575. Development Type
+   576. Vocational Trade
+   577. Artwork Title
+   578. Folklore Form
+   579. Ritual Type
+   580. Device Capability
+   581. Treatment Type
+   582. Ad Format
+   583. Maize Cultivar
+   584. Sacred Site Name
+   585. Leisure Amenity
+   586. Student Type
+   587. Loyalty Program Feature
+   588. Art Installation Type
+   589. Plantation Company
+   590. Damage Nature
+   591. Server Role
+   592. Subsidy Program Name
+   593. RPG System
+   594. Emotion Type
+   595. App Function
+   596. Freight Mode
+   597. Customer Tier
+   598. Art Tradition
+   599. Action Type
+   600. Market Segment
+   601. Narrative Trope
+   602. Home Improvement Type
+   603. Debt Presence
+   604. Ticket Category
+   605. User Generation
+   606. Lifecycle Status
+   607. Playoff Round
+   608. Donor Classification
+   609. Collision Type
+
+📖 读取和聚合数据...
+✅ 总共 1099 个不同的 domains
+
+🔍 过滤后保留 609 个 domains:
+   - US State: 50 values, total count: 57,803
+   - Gender Demographic: 2 values, total count: 11,466
+   - Sport Type: 52 values, total count: 23,756
+   - Entertainment Genre: 19 values, total count: 24,732
+   - Digital Platform: 72 values, total count: 17,662
+   - Trade Direction: 2 values, total count: 4,266
+   - School Category: 6 values, total count: 4,716
+   - Economic System: 2 values, total count: 4,034
+   - Occurrence Frequency: 3 values, total count: 4,035
+   - Canadian Province: 10 values, total count: 9,192
+   - Industry Sector: 136 values, total count: 11,813
+   - Smart Home Category: 5 values, total count: 2,786
+   - Income Bracket: 3 values, total count: 3,501
+   - Monetization Method: 5 values, total count: 1,578
+   - Browser Type: 6 values, total count: 6,242
+   - Media Type: 15 values, total count: 2,681
+   - Product Category: 236 values, total count: 10,987
+   - Device Type: 31 values, total count: 5,665
+   - Transportation Mode: 20 values, total count: 3,910
+   - Art Medium: 25 values, total count: 4,653
+   - Crop Type: 84 values, total count: 8,647
+   - Continent: 7 values, total count: 4,206
+   - Land Use Type: 34 values, total count: 2,488
+   - Sports League: 6 values, total count: 2,962
+   - Time Off Type: 1 values, total count: 752
+   - Political Party: 10 values, total count: 1,961
+   - Socioeconomic Status Level: 3 values, total count: 2,098
+   - Farm Size Category: 3 values, total count: 2,082
+   - Specialization Area: 65 values, total count: 4,396
+   - Calendar Month: 12 values, total count: 2,119
+   - Smartphone Brand: 8 values, total count: 862
+   - Operation Type: 3 values, total count: 1,290
+   - Energy Technology: 73 values, total count: 4,387
+   - Vehicle Classification: 20 values, total count: 2,411
+   - Seaport: 5 values, total count: 1,216
+   - Dog Breed: 25 values, total count: 4,251
+   - Irrigation Method: 16 values, total count: 2,168
+   - Farming System: 4 values, total count: 1,148
+   - TV Network: 7 values, total count: 1,841
+   - Energy Source: 26 values, total count: 4,164
+   - Geographic Classification: 253 values, total count: 7,141
+   - Diplomatic Mission Type: 3 values, total count: 1,386
+   - Material: 75 values, total count: 4,498
+   - Extreme Temperature Type: 4 values, total count: 772
+   - Political Alignment: 7 values, total count: 1,379
+   - Academic Subject: 41 values, total count: 2,599
+   - Pet Category: 3 values, total count: 750
+   - Infrastructure Type: 5 values, total count: 1,015
+   - Product Brand: 38 values, total count: 1,614
+   - Client Engagement Channel: 8 values, total count: 1,197
+   - Physical Activity: 25 values, total count: 1,838
+   - Airport Code: 11 values, total count: 1,335
+   - Country: 41 values, total count: 4,019
+   - Dance Discipline: 16 values, total count: 1,821
+   - Beer Style: 14 values, total count: 1,390
+   - Regulatory Domain: 16 values, total count: 693
+   - Sports Team: 199 values, total count: 6,877
+   - Gemstone Type: 16 values, total count: 1,703
+   - Airline Name: 17 values, total count: 1,521
+   - Audience Segment: 21 values, total count: 845
+   - Clothing Type: 16 values, total count: 1,517
+   - Tea Type: 10 values, total count: 892
+   - Musical Instrument: 32 values, total count: 1,730
+   - Incident Type: 8 values, total count: 842
+   - Building Classification: 19 values, total count: 1,693
+   - Home Appliance: 11 values, total count: 1,235
+   - Medal Type: 3 values, total count: 750
+   - Company Size Category: 3 values, total count: 523
+   - Occupation: 13 values, total count: 845
+   - Soil Texture: 2 values, total count: 414
+   - Calendar Season: 4 values, total count: 844
+   - Company Name: 145 values, total count: 3,551
+   - Athletic Conference: 8 values, total count: 1,001
+   - Demographic Group: 10 values, total count: 613
+   - Cuisine: 21 values, total count: 1,450
+   - Media Asset Type: 7 values, total count: 607
+   - Musical Title: 4 values, total count: 302
+   - Administrative Borough: 5 values, total count: 932
+   - Fabric Type: 8 values, total count: 783
+   - Comedy Type: 3 values, total count: 269
+   - Animal Species: 247 values, total count: 6,478
+   - Performance Type: 6 values, total count: 482
+   - Travel Location: 70 values, total count: 3,717
+   - Phone Type: 2 values, total count: 308
+   - Event Classification: 14 values, total count: 922
+   - Art Period: 6 values, total count: 418
+   - Vehicle Powertrain: 3 values, total count: 386
+   - Digital Service Provider: 40 values, total count: 1,980
+   - Property Type: 10 values, total count: 422
+   - Tillage System: 9 values, total count: 824
+   - Integrity Standing: 11 values, total count: 996
+   - Proficiency Level: 3 values, total count: 329
+   - Opinion Type: 5 values, total count: 750
+   - Superhero: 15 values, total count: 1,075
+   - Streaming Service: 9 values, total count: 689
+   - Operating System: 3 values, total count: 204
+   - Religious Affiliation: 10 values, total count: 599
+   - Weather Type: 9 values, total count: 587
+   - Cooking Technique: 7 values, total count: 610
+   - Flight Disruption Type: 4 values, total count: 304
+   - Casualty Status: 2 values, total count: 309
+   - Subject Area: 14 values, total count: 843
+   - Household Life Stage: 2 values, total count: 189
+   - Animal Adoption Source: 2 values, total count: 171
+   - Disease Name: 15 values, total count: 1,049
+   - Podcast Genre: 10 values, total count: 761
+   - Grant Active: 1 values, total count: 151
+   - Storage Category: 1 values, total count: 151
+   - Family Structure Type: 6 values, total count: 493
+   - Exhibit Theme: 16 values, total count: 596
+   - Biome Type: 14 values, total count: 925
+   - Organization Type: 27 values, total count: 1,020
+   - Football Club: 27 values, total count: 1,472
+   - Craft Category: 9 values, total count: 461
+   - Retail Environment: 5 values, total count: 257
+   - Temperature Record Type: 2 values, total count: 220
+   - Marital Status: 8 values, total count: 348
+   - Playground Feature: 11 values, total count: 722
+   - Creative Artist: 42 values, total count: 936
+   - Nuclear Reactor Type: 3 values, total count: 228
+   - Central Bank Name: 4 values, total count: 404
+   - Art Technique: 20 values, total count: 790
+   - Road Classification: 9 values, total count: 401
+   - Manufacturing Stage: 2 values, total count: 259
+   - Nutritional Component: 9 values, total count: 461
+   - Park Category: 5 values, total count: 326
+   - Island Name: 22 values, total count: 767
+   - Bird Species: 20 values, total count: 600
+   - Minority Indicator: 2 values, total count: 246
+   - Camera Form Factor: 6 values, total count: 308
+   - Radio Genre: 7 values, total count: 420
+   - Risk Rating: 3 values, total count: 331
+   - Venue Type: 7 values, total count: 451
+   - Artifact Class: 14 values, total count: 584
+   - Leisure Activity: 23 values, total count: 877
+   - Toy Type: 23 values, total count: 956
+   - Day Classification: 6 values, total count: 344
+   - Business Segment: 6 values, total count: 415
+   - Pollution Cause Type: 10 values, total count: 434
+   - Home Space Type: 10 values, total count: 364
+   - Attraction Park Name: 49 values, total count: 1,169
+   - Brewing Method: 10 values, total count: 541
+   - Dietary Consideration: 6 values, total count: 399
+   - Proximity Level: 1 values, total count: 110
+   - Medical Injury Type: 5 values, total count: 335
+   - Immigration Status: 1 values, total count: 108
+   - Protein Type: 6 values, total count: 313
+   - Political Entity: 3 values, total count: 133
+   - Revenue Type: 6 values, total count: 227
+   - Video Game Title: 17 values, total count: 722
+   - Social Event Type: 5 values, total count: 217
+   - Purchase Channel: 4 values, total count: 316
+   - Harvest Method: 2 values, total count: 194
+   - Investment Asset Class: 5 values, total count: 329
+   - Family Member Count: 3 values, total count: 269
+   - Involvement Level: 2 values, total count: 191
+   - Payment Method: 11 values, total count: 331
+   - Fertilizer Type: 6 values, total count: 175
+   - Dwelling Type: 15 values, total count: 524
+   - Music Genre: 13 values, total count: 454
+   - Expense Type: 23 values, total count: 863
+   - Climate Zone Type: 4 values, total count: 168
+   - App Pricing Model: 1 values, total count: 90
+   - Vehicle Model: 28 values, total count: 678
+   - Ecosystem Type: 9 values, total count: 365
+   - Education Format: 2 values, total count: 174
+   - City: 21 values, total count: 517
+   - Medical Procedure: 12 values, total count: 425
+   - Produce Name: 11 values, total count: 344
+   - Landmark Name: 43 values, total count: 860
+   - Day Period: 4 values, total count: 264
+   - Building Style: 11 values, total count: 347
+   - Lighting Type: 3 values, total count: 134
+   - River Basin Name: 11 values, total count: 376
+   - Sales Metric: 1 values, total count: 80
+   - Recipient Category: 5 values, total count: 247
+   - Sport Position: 3 values, total count: 125
+   - Cover Crop Type: 15 values, total count: 403
+   - Generational Role: 6 values, total count: 300
+   - Turf Type: 2 values, total count: 156
+   - Occupational Role: 44 values, total count: 1,022
+   - Seafood Species: 31 values, total count: 896
+   - Metal Type: 5 values, total count: 267
+   - Marine Conservation Name: 17 values, total count: 451
+   - Protected Area Name: 37 values, total count: 807
+   - Travel Category: 6 values, total count: 308
+   - Career Level: 4 values, total count: 203
+   - Food Category: 7 values, total count: 301
+   - Tourism Type: 2 values, total count: 143
+   - Pandemic Period: 2 values, total count: 114
+   - Color: 14 values, total count: 507
+   - Galaxy Type: 4 values, total count: 225
+   - Air Quality Parameter: 5 values, total count: 186
+   - Social Media Intensity: 2 values, total count: 144
+   - Engagement: 2 values, total count: 144
+   - Cryptocurrency Name: 2 values, total count: 82
+   - Accessory Category: 9 values, total count: 372
+   - Launch Status: 2 values, total count: 142
+   - Jewelry Design Style: 8 values, total count: 371
+   - National Park Name: 10 values, total count: 287
+   - Legislative Status: 2 values, total count: 116
+   - Show Title: 8 values, total count: 286
+   - Art Movement: 4 values, total count: 119
+   - Health Coverage Type: 2 values, total count: 130
+   - Interest Category: 7 values, total count: 272
+   - Currency Code: 3 values, total count: 105
+   - Lake Name: 15 values, total count: 379
+   - Plant Establishment Method: 2 values, total count: 123
+   - Health Area: 4 values, total count: 175
+   - Mobile OS: 3 values, total count: 139
+   - Financial Incentive Type: 2 values, total count: 114
+   - Affectionate Gesture: 2 values, total count: 73
+   - Crypto Exchange: 6 values, total count: 255
+   - Higher Education Type: 2 values, total count: 119
+   - Medical Specialty: 22 values, total count: 592
+   - Eating Occasion: 4 values, total count: 206
+   - Music Subgenre: 6 values, total count: 202
+   - Storage Type: 4 values, total count: 118
+   - Battleground State: 2 values, total count: 76
+   - Zoning District: 6 values, total count: 207
+   - Data Collection Method: 6 values, total count: 130
+   - Season Phase: 2 values, total count: 93
+   - Phone Model: 16 values, total count: 418
+   - Environmental Impact: 7 values, total count: 224
+   - Coffee Style: 6 values, total count: 150
+   - Education Category: 17 values, total count: 330
+   - Game Name: 30 values, total count: 674
+   - Holiday Name: 8 values, total count: 251
+   - Waste Material: 4 values, total count: 164
+   - Museum Focus: 3 values, total count: 139
+   - Cargo Type: 3 values, total count: 124
+   - Game Genre: 8 values, total count: 228
+   - Acquisition Channel: 3 values, total count: 87
+   - Louisiana Parish: 8 values, total count: 242
+   - Conservation Status: 4 values, total count: 175
+   - Agricultural Practice: 16 values, total count: 370
+   - Funding Source: 4 values, total count: 142
+   - Supply Chain Stage: 11 values, total count: 271
+   - Application Name: 1 values, total count: 52
+   - Engineering Structure Type: 4 values, total count: 116
+   - Recipe Ingredient: 11 values, total count: 308
+   - Sofa Type: 3 values, total count: 110
+   - Disruption Reason: 4 values, total count: 84
+   - Financial Product Type: 6 values, total count: 226
+   - Festival Theme: 6 values, total count: 170
+   - Camera Lens Type: 2 values, total count: 98
+   - Sales Offer Type: 3 values, total count: 92
+   - Concession Product: 5 values, total count: 151
+   - Primate Species: 5 values, total count: 144
+   - Climate Type: 7 values, total count: 188
+   - Listening Method: 4 values, total count: 120
+   - Sports Facility Type: 7 values, total count: 236
+   - Performance Standing: 2 values, total count: 75
+   - Mission Complexity Level: 3 values, total count: 140
+   - Innovation Area: 4 values, total count: 129
+   - Parental Involvement Level: 2 values, total count: 76
+   - Sustainability Category: 5 values, total count: 96
+   - Travel Segment: 3 values, total count: 137
+   - Grain Category: 7 values, total count: 144
+   - Enforcement Degree: 3 values, total count: 138
+   - Fitness Goal: 5 values, total count: 134
+   - Light Pollution Severity: 2 values, total count: 86
+   - Social Determinant: 5 values, total count: 120
+   - Interaction Channel: 3 values, total count: 76
+   - Probability Basis: 2 values, total count: 58
+   - Medical Condition: 8 values, total count: 246
+   - Tourist Attraction: 8 values, total count: 190
+   - Sanction Measure: 4 values, total count: 106
+   - Furniture Category: 8 values, total count: 220
+   - Beverage Type: 18 values, total count: 398
+   - Education Level: 3 values, total count: 121
+   - Heritage Site Type: 9 values, total count: 179
+   - Earring Type: 5 values, total count: 179
+   - Animal Type: 6 values, total count: 172
+   - Achievement Award Category: 6 values, total count: 167
+   - Drug Name: 5 values, total count: 127
+   - Photography Genre: 4 values, total count: 126
+   - Developmental Skill Category: 2 values, total count: 83
+   - Alcoholic Beverage Type: 2 values, total count: 78
+   - Research Publication Type: 3 values, total count: 72
+   - Feedstock Type: 5 values, total count: 150
+   - Flyway Region: 4 values, total count: 138
+   - Usage Scenario: 6 values, total count: 137
+   - Adoption Level: 2 values, total count: 82
+   - Debris Material: 4 values, total count: 117
+   - Content Section: 4 values, total count: 110
+   - Design Style: 3 values, total count: 61
+   - Establishment Type: 6 values, total count: 152
+   - Membership Category: 4 values, total count: 136
+   - Satellite Function: 3 values, total count: 93
+   - Building Siding Material: 2 values, total count: 65
+   - Propulsion Type: 10 values, total count: 231
+   - Health Condition Category: 4 values, total count: 104
+   - Property Flooring Type: 2 values, total count: 78
+   - Connected Service: 4 values, total count: 75
+   - Fantasy Entity: 6 values, total count: 190
+   - Capture Method: 7 values, total count: 171
+   - Outreach Channel: 7 values, total count: 164
+   - Store Category: 6 values, total count: 127
+   - Marketing Channel: 6 values, total count: 123
+   - Data Domain: 7 values, total count: 146
+   - Art Subject: 3 values, total count: 104
+   - Cyber Attack Type: 3 values, total count: 75
+   - IP Infringement Type: 3 values, total count: 73
+   - Milestone Type: 2 values, total count: 54
+   - Medicine System: 2 values, total count: 48
+   - Major Sports Event: 8 values, total count: 197
+   - Citrus Variety: 5 values, total count: 151
+   - Insurance Type: 5 values, total count: 150
+   - Food Item: 4 values, total count: 129
+   - Award Ceremony: 7 values, total count: 146
+   - Tomato Variety: 6 values, total count: 126
+   - Mitigation Method: 9 values, total count: 152
+   - Insect Common Name: 3 values, total count: 99
+   - Driving Context: 3 values, total count: 91
+   - Funding Type: 4 values, total count: 84
+   - Price Level: 2 values, total count: 74
+   - Venue Seating Section: 1 values, total count: 37
+   - Cancer Type: 6 values, total count: 161
+   - Home Decor Category: 9 values, total count: 155
+   - Application Category: 6 values, total count: 148
+   - Music Distribution Format: 5 values, total count: 145
+   - Wearable Item Type: 5 values, total count: 127
+   - Court Surface: 3 values, total count: 96
+   - Play Type: 10 values, total count: 207
+   - Reef Site: 8 values, total count: 163
+   - Gemstone Color: 7 values, total count: 191
+   - Lighting Fixture Type: 6 values, total count: 148
+   - Livestock Category: 4 values, total count: 88
+   - Cultivation Environment: 2 values, total count: 70
+   - Sport Tier: 3 values, total count: 65
+   - Blockchain Platform: 1 values, total count: 35
+   - Wine Classification: 8 values, total count: 207
+   - Participation Level: 8 values, total count: 170
+   - Exercise Type: 7 values, total count: 162
+   - Content Streaming Platform: 8 values, total count: 158
+   - Email Provider: 5 values, total count: 142
+   - Sensor Category: 7 values, total count: 124
+   - Student Proficiency: 3 values, total count: 102
+   - Costume Character: 4 values, total count: 101
+   - Manure Management Method: 3 values, total count: 68
+   - Class Classification: 15 values, total count: 313
+   - Historical Period: 12 values, total count: 267
+   - Vineyard Designation: 7 values, total count: 134
+   - Care Level: 3 values, total count: 88
+   - Religious Symbol: 3 values, total count: 85
+   - Manufacturing Step: 5 values, total count: 85
+   - Pollution Source: 3 values, total count: 81
+   - Permit Sector: 3 values, total count: 66
+   - Cost Element: 2 values, total count: 56
+   - Fruit Name: 12 values, total count: 220
+   - Neighborhood: 11 values, total count: 214
+   - Amusement Ride Type: 7 values, total count: 142
+   - Dietary Food Group: 4 values, total count: 89
+   - Agricultural Region: 3 values, total count: 63
+   - Narrative Device: 2 values, total count: 46
+   - Gaming Publication: 7 values, total count: 173
+   - Economic Region: 8 values, total count: 132
+   - Retailer Name: 7 values, total count: 117
+   - Pesticide Classification: 4 values, total count: 116
+   - Dish Type: 7 values, total count: 114
+   - Impact Area: 5 values, total count: 114
+   - Painting Medium: 4 values, total count: 107
+   - Vehicle Maintenance Type: 5 values, total count: 104
+   - Seabird Species: 6 values, total count: 100
+   - Professional Skill: 6 values, total count: 91
+   - Campaign Type: 4 values, total count: 88
+   - Remediation Method: 3 values, total count: 69
+   - Insulation Material: 5 values, total count: 105
+   - Device Capability: 1 values, total count: 31
+   - Performance Indicator: 13 values, total count: 238
+   - Craft Supply: 11 values, total count: 224
+   - Land Management Practice: 6 values, total count: 149
+   - Vessel Classification: 8 values, total count: 141
+   - Shipping Passage: 4 values, total count: 112
+   - Forest Cover Type: 5 values, total count: 106
+   - Comic Issue: 5 values, total count: 97
+   - Geographic Region: 8 values, total count: 153
+   - GDP Component Type: 4 values, total count: 91
+   - Shot Zone: 5 values, total count: 88
+   - Irrigation Source Type: 3 values, total count: 75
+   - Project Phase: 3 values, total count: 69
+   - Sport Gender Division: 2 values, total count: 60
+   - Water Supply Type: 2 values, total count: 60
+   - Video Platform: 2 values, total count: 55
+   - Index Basis: 2 values, total count: 54
+   - Dress Silhouette: 2 values, total count: 42
+   - Sacred Site Name: 1 values, total count: 30
+   - Gift Category: 6 values, total count: 107
+   - Gallery Sector: 2 values, total count: 53
+   - App Usage Level: 3 values, total count: 85
+   - Hurricane Wind Category: 4 values, total count: 82
+   - Mitigation Project Type: 3 values, total count: 73
+   - Agricultural Type: 6 values, total count: 119
+   - Hub City: 2 values, total count: 58
+   - Internet Access Status: 2 values, total count: 58
+   - Tariff Active: 2 values, total count: 58
+   - Patrol Method: 2 values, total count: 52
+   - Reef Region: 3 values, total count: 49
+   - Performance Terrain: 2 values, total count: 42
+   - Pest Category: 11 values, total count: 235
+   - Measure Type: 5 values, total count: 118
+   - Workout Modality: 4 values, total count: 94
+   - Impact Severity: 2 values, total count: 56
+   - Sustainable Feature: 3 values, total count: 84
+   - Green Space Accessibility: 3 values, total count: 84
+   - Library Name: 2 values, total count: 50
+   - Subscription Action Type: 2 values, total count: 38
+   - Customer Segment Name: 16 values, total count: 254
+   - Fatal Incident Type: 8 values, total count: 141
+   - Tractor Configuration: 4 values, total count: 104
+   - Service Category: 13 values, total count: 210
+   - Accessibility Feature Type: 5 values, total count: 99
+   - Consumer Profile: 4 values, total count: 96
+   - Resource Classification: 3 values, total count: 78
+   - Manufacturing Process: 4 values, total count: 74
+   - Financial Concern Category: 3 values, total count: 69
+   - Control Technique: 3 values, total count: 59
+   - Weightlifting Exercise: 6 values, total count: 135
+   - Software Application: 5 values, total count: 106
+   - Support Category: 4 values, total count: 81
+   - Therapeutic Area: 7 values, total count: 126
+   - Orchard Region: 3 values, total count: 78
+   - Access Level: 2 values, total count: 52
+   - Result Status: 2 values, total count: 52
+   - Subsidy Indicator: 2 values, total count: 52
+   - Participation Status: 4 values, total count: 74
+   - Japan Prefecture: 2 values, total count: 46
+   - Worker Type: 2 values, total count: 37
+   - Disaster Event: 10 values, total count: 197
+   - Event Category: 5 values, total count: 101
+   - Packaging Material: 7 values, total count: 125
+   - Filtration Technology: 4 values, total count: 65
+   - Election Type: 3 values, total count: 55
+   - Wellness Program Type: 3 values, total count: 51
+   - Marine Area Name: 11 values, total count: 193
+   - Public Policy: 6 values, total count: 124
+   - Market Trend: 9 values, total count: 139
+   - Military Platform Type: 8 values, total count: 124
+   - ADAS Module: 6 values, total count: 118
+   - Rover Name: 5 values, total count: 104
+   - Organization Sector: 4 values, total count: 75
+   - Ballet Title: 4 values, total count: 75
+   - Subscription Plan: 3 values, total count: 52
+   - Footwear Style: 3 values, total count: 72
+   - Environmental Strategy: 4 values, total count: 71
+   - Space Mission: 3 values, total count: 64
+   - National Leader: 2 values, total count: 48
+   - Aircraft Model: 11 values, total count: 198
+   - Project Category: 7 values, total count: 108
+   - News Organization: 13 values, total count: 184
+   - Zone Designation: 4 values, total count: 62
+   - Pruning Style: 3 values, total count: 61
+   - Clinical Service: 3 values, total count: 59
+   - Hat Style: 3 values, total count: 58
+   - Forage Type: 3 values, total count: 53
+   - Ranching Approach: 2 values, total count: 33
+   - Corridor Name: 8 values, total count: 110
+   - Project Type: 7 values, total count: 121
+   - Service Branch: 5 values, total count: 100
+   - Source Collection: 5 values, total count: 86
+   - App Permission: 5 values, total count: 86
+   - Production Type: 5 values, total count: 82
+   - Home Feature: 4 values, total count: 77
+   - Booking Point: 4 values, total count: 76
+   - Education Focus: 4 values, total count: 67
+   - Loan Application Status: 3 values, total count: 54
+   - Ecological Area Type: 3 values, total count: 53
+   - Educational Attainment: 3 values, total count: 51
+   - Pricing Basis: 2 values, total count: 44
+   - Route Type: 2 values, total count: 37
+   - System Component: 12 values, total count: 174
+   - Conservation Organization: 9 values, total count: 146
+   - Aid Sector: 5 values, total count: 105
+   - Flavor Profile: 4 values, total count: 73
+   - Essential Need: 3 values, total count: 63
+   - Craft Material: 4 values, total count: 63
+   - Fashion Style: 8 values, total count: 125
+   - Character Kind: 3 values, total count: 45
+   - Aircraft Category: 2 values, total count: 42
+   - Flag State Name: 5 values, total count: 100
+   - Roofing Type: 6 values, total count: 98
+   - Technology Solution: 8 values, total count: 96
+   - Fishery Location: 5 values, total count: 84
+   - Amphibian Species Name: 5 values, total count: 81
+   - Ballpark Name: 6 values, total count: 78
+   - Tributary Name: 4 values, total count: 71
+   - Instructional Method: 5 values, total count: 76
+   - Mammal Species Name: 3 values, total count: 60
+   - Insect Type: 3 values, total count: 60
+   - Transit Station: 5 values, total count: 70
+   - Theater Format: 2 values, total count: 40
+   - Love Language Type: 2 values, total count: 40
+   - Pollinator Classification: 2 values, total count: 40
+   - Treatment Type: 2 values, total count: 31
+   - Food Product Name: 10 values, total count: 134
+   - Work Gear Type: 5 values, total count: 87
+   - Space Agency: 4 values, total count: 76
+   - Microphone Classification: 4 values, total count: 70
+   - Management Technique: 4 values, total count: 63
+   - Product Form: 4 values, total count: 60
+   - Washing Machine Form Factor: 4 values, total count: 56
+   - Subject Category: 3 values, total count: 50
+   - Conflict Country: 3 values, total count: 47
+   - Artwork Title: 2 values, total count: 32
+   - Employment Model: 2 values, total count: 36
+   - Computer Peripheral: 4 values, total count: 57
+   - Farm Activity: 3 values, total count: 54
+   - Agricultural Pest: 3 values, total count: 51
+   - Federal Agency: 2 values, total count: 35

icon_generation/backup/extract_accepted_domains.py

@@ -0,0 +1,171 @@
+#!/usr/bin/env python3
+"""
+从 domain_value_pairs_enhanced.txt 中提取 accepted_domains.txt 中的 domains
+并按指定格式组织输出
+"""
+
+from collections import defaultdict
+from typing import Set, Dict, List
+
+
+def parse_csv_line(line: str) -> tuple:
+    """解析 CSV 行"""
+    parts = []
+    current = []
+    in_quotes = False
+    
+    for char in line:
+        if char == '"':
+            in_quotes = not in_quotes
+        elif char == ',' and not in_quotes:
+            parts.append(''.join(current).strip())
+            current = []
+        else:
+            current.append(char)
+    parts.append(''.join(current).strip())
+    
+    if len(parts) >= 3:
+        domain = parts[0]
+        attribute = parts[1]
+        freq = int(parts[2])
+        return domain, attribute, freq
+    else:
+        return None, None, 0
+
+
+def load_accepted_domains(file_path: str) -> Set[str]:
+    """加载可接受的 domains"""
+    domains = set()
+    with open(file_path, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if line:
+                domains.add(line)
+    return domains
+
+
+def extract_accepted_domains(
+    input_file: str,
+    accepted_domains_file: str,
+    output_file: str = None
+):
+    """
+    提取 accepted domains 的 domain-attribute pairs
+    
+    Args:
+        input_file: 输入的 domain_value_pairs_enhanced.txt
+        accepted_domains_file: accepted_domains.txt
+        output_file: 输出文件路径（可选）
+    """
+    print("=" * 80)
+    print("提取 Accepted Domains 的 Domain-Attribute Pairs")
+    print("=" * 80)
+    print()
+    
+    # 1. 加载可接受的 domains
+    print(f"📖 读取可接受的 domains: {accepted_domains_file}")
+    accepted_domains = load_accepted_domains(accepted_domains_file)
+    print(f"✅ 可接受的 domains: {len(accepted_domains)} 个")
+    print()
+    
+    # 2. 读取并过滤 domain-attribute pairs
+    print(f"📖 读取 domain-attribute pairs: {input_file}")
+    domain_attributes = defaultdict(list)
+    
+    with open(input_file, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                continue
+            
+            domain, attribute, freq = parse_csv_line(line)
+            if domain and domain in accepted_domains:
+                domain_attributes[domain].append((attribute, freq))
+    
+    print(f"✅ 找到 {len(domain_attributes)} 个匹配的 domains")
+    print()
+    
+    # 3. 对每个 domain 的 attributes 按 frequency 排序
+    for domain in domain_attributes:
+        domain_attributes[domain].sort(key=lambda x: x[1], reverse=True)
+    
+    # 4. 按 accepted_domains.txt 的顺序排序 domains
+    # 保持 accepted_domains.txt 中的顺序
+    accepted_list = []
+    with open(accepted_domains_file, 'r', encoding='utf-8') as f:
+        for line in f:
+            domain = line.strip()
+            if domain and domain in domain_attributes:
+                accepted_list.append(domain)
+    
+    # 添加任何在数据中但不在列表中的 domains（以防万一）
+    for domain in domain_attributes:
+        if domain not in accepted_list:
+            accepted_list.append(domain)
+    
+    # 5. 生成输出
+    print("=" * 80)
+    print("📊 生成的 Domain-Attribute 列表:")
+    print("=" * 80)
+    print()
+    
+    output_lines = []
+    
+    for domain in accepted_list:
+        if domain not in domain_attributes:
+            continue
+        
+        attributes = domain_attributes[domain]
+        attr_names = [attr for attr, _ in attributes]
+        attr_str = ", ".join(attr_names)
+        
+        # 添加到输出
+        output_lines.append(domain)
+        output_lines.append(attr_str)
+        output_lines.append("")  # 空行
+        
+        # 打印到控制台
+        print(domain)
+        print(attr_str)
+        print()
+    
+    # 6. 保存到文件（如果指定）
+    if output_file:
+        print("=" * 80)
+        print(f"💾 保存到文件: {output_file}")
+        
+        with open(output_file, 'w', encoding='utf-8') as f:
+            for line in output_lines:
+                f.write(line + "\n")
+        
+        print(f"✅ 已保存")
+    
+    # 7. 统计信息
+    print()
+    print("=" * 80)
+    print("📈 统计信息:")
+    print("=" * 80)
+    print(f"匹配的 Domains 数: {len(domain_attributes)}")
+    print(f"总 Attributes 数: {sum(len(attrs) for attrs in domain_attributes.values())}")
+    
+    # 显示每个 domain 的 attributes 数量
+    print()
+    print("各 Domain 的 Attributes 数量:")
+    for domain in accepted_list:
+        if domain in domain_attributes:
+            count = len(domain_attributes[domain])
+            print(f"   {domain}: {count} attributes")
+    
+    print()
+    print("=" * 80)
+    print("✨ 提取完成！")
+    print("=" * 80)
+
+
+if __name__ == '__main__':
+    input_file = 'domain_value_pairs_enhanced.txt'
+    accepted_domains_file = 'accepted_domains.txt'
+    output_file = 'accepted_domains_attributes.txt'
+    
+    extract_accepted_domains(input_file, accepted_domains_file, output_file)
+

icon_generation/backup/filter_accepted_domains.py

@@ -0,0 +1,445 @@
+#!/usr/bin/env python3
+"""
+过滤和精简 accepted_domains_attributes.txt
+1. 删除过于小众的 domains
+2. 对于 attributes 不容易区分的 domain，只保留 5-10 个最重要的
+3. 大多数正常的 domain 完全保留
+"""
+
+import json
+import os
+import requests
+from typing import Dict, Optional, List, Tuple
+from concurrent.futures import ThreadPoolExecutor, as_completed
+import threading
+import time
+
+
+class DomainFilter:
+    """使用 LLM 过滤和精简 domains"""
+    
+    def __init__(self, api_key=None, base_url=None, model=None):
+        """初始化 LLM analyzer"""
+        self.api_key = api_key or os.getenv("OPENAI_API_KEY") or os.getenv("AIHUBMIX_API_KEY", "")
+        self.base_url = base_url or os.getenv("OPENAI_BASE_URL", "https://aihubmix.com/v1")
+        self.model = model or os.getenv("OPENAI_MODEL", "gpt-5.2")
+        self.lock = threading.Lock()
+        self.completed_count = 0
+        self.total_domains = 0
+    
+    def filter_domains_batch(self, domains_data: List[Tuple[str, List[str]]], idx: int, total: int, start_time: float) -> Dict:
+        """
+        批量处理 domains（每次 10 个）
+        
+        Args:
+            domains_data: [(domain, [attributes]), ...] 最多 10 个
+            idx: 批次索引
+            total: 总批次数
+            start_time: 开始时间
+        
+        Returns:
+            {
+                'keep': [(domain, [attributes]), ...],
+                'remove': [domain, ...],
+                'reasoning': 'explanation'
+            }
+        """
+        prompt = self._build_filter_prompt(domains_data)
+        
+        with self.lock:
+            progress = (idx / total) * 100
+            elapsed = time.time() - start_time
+            avg_time = elapsed / idx if idx > 0 else 0
+            remaining = avg_time * (total - idx)
+            
+            print(f"\n{'=' * 80}")
+            print(f"🔄 处理批次 {idx}/{total} ({progress:.1f}%)")
+            print(f"⏱️  已用时间: {elapsed:.1f}秒 | 预计剩余: {remaining:.1f}秒")
+            print(f"   当前批次: {len(domains_data)} 个 domains")
+            print(f"   🤖 调用 LLM 进行过滤和精简...")
+        
+        try:
+            response = self._query_llm(prompt)
+            
+            if response:
+                # 清理可能的 markdown 代码块
+                cleaned_response = response.strip()
+                if cleaned_response.startswith('```'):
+                    lines = cleaned_response.split('\n')
+                    cleaned_response = '\n'.join(lines[1:-1] if lines[-1].strip() == '```' else lines[1:])
+                    cleaned_response = cleaned_response.replace('```json', '').replace('```', '').strip()
+                
+                result = json.loads(cleaned_response)
+                
+                # 验证返回格式
+                if 'domains' in result:
+                    keep = []
+                    remove = []
+                    
+                    for item in result['domains']:
+                        domain = item['domain']
+                        action = item.get('action', 'keep')
+                        
+                        if action == 'remove':
+                            remove.append(domain)
+                        elif action == 'keep_all':
+                            # 找到原始数据
+                            for orig_domain, orig_attrs in domains_data:
+                                if orig_domain == domain:
+                                    keep.append((domain, orig_attrs))
+                                    break
+                        elif action == 'keep_reduced':
+                            # 使用精简后的 attributes
+                            reduced_attrs = item.get('reduced_attributes', [])
+                            if reduced_attrs:
+                                keep.append((domain, reduced_attrs))
+                            else:
+                                # 如果没有提供，保留原始
+                                for orig_domain, orig_attrs in domains_data:
+                                    if orig_domain == domain:
+                                        keep.append((domain, orig_attrs))
+                                        break
+                        else:
+                            # 默认保留
+                            for orig_domain, orig_attrs in domains_data:
+                                if orig_domain == domain:
+                                    keep.append((domain, orig_attrs))
+                                    break
+                    
+                    with self.lock:
+                        self.completed_count += 1
+                        print(f"   ✅ 处理完成!")
+                        print(f"      保留: {len(keep)} 个")
+                        print(f"      删除: {len(remove)} 个")
+                        if remove:
+                            print(f"      删除的 domains: {', '.join(remove)}")
+                    
+                    return {
+                        'keep': keep,
+                        'remove': remove,
+                        'reasoning': result.get('reasoning', '')
+                    }
+                else:
+                    with self.lock:
+                        self.completed_count += 1
+                        print(f"      ⚠️  LLM 响应缺少必需字段，保留所有")
+                    # 默认全部保留
+                    return {
+                        'keep': [(domain, attrs) for domain, attrs in domains_data],
+                        'remove': [],
+                        'reasoning': 'LLM response format error, kept all'
+                    }
+                    
+            else:
+                with self.lock:
+                    self.completed_count += 1
+                    print(f"      ⚠️  LLM API 调用失败，保留所有")
+                return {
+                    'keep': [(domain, attrs) for domain, attrs in domains_data],
+                    'remove': [],
+                    'reasoning': 'LLM API failed, kept all'
+                }
+                
+        except json.JSONDecodeError as e:
+            with self.lock:
+                self.completed_count += 1
+                print(f"      ⚠️  LLM 响应不是有效的 JSON: {e}")
+            return {
+                'keep': [(domain, attrs) for domain, attrs in domains_data],
+                'remove': [],
+                'reasoning': 'JSON decode error, kept all'
+            }
+        except Exception as e:
+            with self.lock:
+                self.completed_count += 1
+                print(f"      ⚠️  处理错误: {e}")
+            return {
+                'keep': [(domain, attrs) for domain, attrs in domains_data],
+                'remove': [],
+                'reasoning': f'Error: {str(e)}, kept all'
+            }
+    
+    def _build_filter_prompt(self, domains_data: List[Tuple[str, List[str]]]) -> str:
+        """构建过滤的 prompt"""
+        
+        domains_str = ""
+        for domain, attributes in domains_data:
+            attrs_str = ", ".join(attributes[:20])  # 最多显示前20个
+            if len(attributes) > 20:
+                attrs_str += f", ... (共 {len(attributes)} 个)"
+            domains_str += f"\nDomain: {domain}\nAttributes: {attrs_str}\n"
+        
+        prompt = f"""You are a data curation expert. Given a list of domains and their attributes, your task is to:
+
+1. **REMOVE** domains that are too niche, specialized, or not commonly used (e.g., "Lemur Taxon Name (Genus and Species)", highly technical scientific classifications, extremely specific subcategories)
+2. **KEEP ALL** domains that are mainstream, widely recognized, and useful for general purposes (e.g., "Industry Sector", "Land Use Type", "Product Category", "Transportation Mode")
+3. **REDUCE** attributes for domains where attributes are hard to distinguish or too granular (e.g., "Donor Classification" with "Individual Donor", "One-time Donor", "Recurring Donor", "Monthly Donor" - these are too similar). For such domains, keep only 5-10 most important and distinct attributes.
+
+Guidelines:
+- **REMOVE** if the domain is:
+  - Too specific or scientific (e.g., species names, technical taxonomies)
+  - Too niche or rarely used in general contexts
+  - Overly granular subcategories
+  
+- **KEEP ALL** if the domain is:
+  - Commonly used in business, data analysis, or general applications
+  - Well-known categories (e.g., industries, product types, locations)
+  - Useful for visualization or categorization
+  
+- **REDUCE** attributes if:
+  - Attributes are very similar or hard to distinguish
+  - Too many granular variations (e.g., "Monthly Donor" vs "Recurring Donor")
+  - Keep 5-10 most important and distinct ones
+
+Domains to evaluate:
+{domains_str}
+
+Return your response in the following JSON format ONLY (no markdown, no extra text):
+{{
+  "domains": [
+    {{
+      "domain": "Domain Name",
+      "action": "remove",
+      "reduced_attributes": []
+    }},
+    {{
+      "domain": "Domain Name",
+      "action": "keep_all",
+      "reduced_attributes": []
+    }},
+    {{
+      "domain": "Domain Name",
+      "action": "keep_reduced",
+      "reduced_attributes": ["attr1", "attr2", "attr3"]
+    }}
+  ],
+  "reasoning": "Brief explanation of decisions"
+}}
+
+Examples:
+- "Lemur Taxon Name (Genus and Species)" → action: "remove" (too niche)
+- "Industry Sector" → action: "keep_all" (mainstream, useful)
+- "Donor Classification" with many similar attributes → action: "keep_reduced", reduced_attributes: ["Individual", "Corporate", "Foundation", "Government", "Anonymous"]
+"""
+        
+        return prompt
+    
+    def _query_llm(self, prompt: str) -> Optional[str]:
+        """查询 LLM API"""
+        headers = {
+            'Authorization': f'Bearer {self.api_key}',
+            'Content-Type': 'application/json'
+        }
+        
+        data = {
+            'model': self.model,
+            'messages': [
+                {
+                    'role': 'system', 
+                    'content': 'You are a data curation expert specialized in filtering and organizing domain-attribute pairs. Always return valid JSON format only, without any markdown formatting or extra text.'
+                },
+                {
+                    'role': 'user', 
+                    'content': prompt
+                }
+            ],
+            'temperature': 0.3
+        }
+        
+        try:
+            response = requests.post(
+                f'{self.base_url}/chat/completions',
+                headers=headers,
+                json=data,
+                timeout=60
+            )
+            response.raise_for_status()
+            
+            result = response.json()
+            return result['choices'][0]['message']['content'].strip()
+            
+        except requests.exceptions.Timeout:
+            with self.lock:
+                print("      ❌ LLM API 超时")
+            return None
+        except requests.exceptions.HTTPError as e:
+            with self.lock:
+                print(f"      ❌ LLM API HTTP 错误: {e}")
+            return None
+        except requests.exceptions.RequestException as e:
+            with self.lock:
+                print(f"      ❌ LLM API 请求错误: {e}")
+            return None
+        except KeyError as e:
+            with self.lock:
+                print(f"      ❌ LLM API 响应格式错误: {e}")
+            return None
+
+
+def load_domains_attributes(file_path: str) -> List[Tuple[str, List[str]]]:
+    """加载 domains 和 attributes"""
+    domains_data = []
+    current_domain = None
+    current_attrs = []
+    
+    with open(file_path, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                # 空行表示一个 domain 结束
+                if current_domain:
+                    domains_data.append((current_domain, current_attrs))
+                    current_domain = None
+                    current_attrs = []
+                continue
+            
+            # 检查是否是新的 domain（没有逗号，且不是 attributes 行）
+            if ',' not in line and not line.startswith(' ') and len(line) > 0:
+                # 保存之前的 domain
+                if current_domain:
+                    domains_data.append((current_domain, current_attrs))
+                # 开始新的 domain
+                current_domain = line
+                current_attrs = []
+            else:
+                # 这是 attributes 行
+                if current_domain:
+                    attrs = [attr.strip() for attr in line.split(',')]
+                    current_attrs.extend(attrs)
+    
+    # 保存最后一个 domain
+    if current_domain:
+        domains_data.append((current_domain, current_attrs))
+    
+    return domains_data
+
+
+def main():
+    print("=" * 80)
+    print("Domain-Attribute 过滤和精简工具")
+    print("=" * 80)
+    print()
+    
+    input_file = 'accepted_domains_attributes.txt'
+    output_file = 'accepted_domains_attributes_filtered.txt'
+    
+    # 1. 加载数据
+    print(f"📖 读取文件: {input_file}")
+    domains_data = load_domains_attributes(input_file)
+    print(f"✅ 成功读取 {len(domains_data)} 个 domains")
+    print()
+    
+    # 2. 初始化过滤器
+    filter_obj = DomainFilter()
+    
+    # 3. 分批处理（每批 10 个）
+    batch_size = 10
+    batches = []
+    for i in range(0, len(domains_data), batch_size):
+        batch = domains_data[i:i+batch_size]
+        batches.append(batch)
+    
+    total_batches = len(batches)
+    filter_obj.total_domains = total_batches
+    
+    print("=" * 80)
+    print(f"🚀 开始处理 (10线程并发)")
+    print(f"   总 domains: {len(domains_data)}")
+    print(f"   批次数: {total_batches}")
+    print(f"   每批: {batch_size} 个 domains")
+    print("=" * 80)
+    print()
+    
+    start_time = time.time()
+    
+    # 4. 使用线程池并发处理
+    filtered_results = []
+    num_threads = 10
+    
+    with ThreadPoolExecutor(max_workers=num_threads) as executor:
+        # 提交所有任务
+        future_to_batch = {
+            executor.submit(
+                filter_obj.filter_domains_batch,
+                batch,
+                idx + 1,
+                total_batches,
+                start_time
+            ): (idx, batch)
+            for idx, batch in enumerate(batches)
+        }
+        
+        # 收集结果
+        for future in as_completed(future_to_batch):
+            idx, batch = future_to_batch[future]
+            try:
+                result = future.result()
+                filtered_results.append(result)
+            except Exception as e:
+                with filter_obj.lock:
+                    print(f"❌ 处理批次 {idx + 1} 时出错: {e}")
+                # 默认保留
+                filtered_results.append({
+                    'keep': [(domain, attrs) for domain, attrs in batch],
+                    'remove': [],
+                    'reasoning': f'Error: {str(e)}'
+                })
+    
+    # 5. 合并结果
+    print()
+    print("=" * 80)
+    print("📊 合并结果...")
+    print("=" * 80)
+    
+    final_domains = []
+    removed_domains = []
+    
+    for result in filtered_results:
+        final_domains.extend(result['keep'])
+        removed_domains.extend(result['remove'])
+    
+    # 6. 保存结果
+    print()
+    print("=" * 80)
+    print("💾 保存结果...")
+    print("=" * 80)
+    
+    with open(output_file, 'w', encoding='utf-8') as f:
+        for domain, attributes in final_domains:
+            f.write(f"{domain}\n")
+            f.write(f"{', '.join(attributes)}\n")
+            f.write("\n")
+    
+    elapsed_time = time.time() - start_time
+    
+    # 7. 统计信息
+    total_attributes = sum(len(attrs) for _, attrs in final_domains)
+    
+    print()
+    print("=" * 80)
+    print("📊 统计信息:")
+    print("=" * 80)
+    print(f"原始 domains 数: {len(domains_data)}")
+    print(f"保留 domains 数: {len(final_domains)}")
+    print(f"删除 domains 数: {len(removed_domains)}")
+    print(f"总 attributes 数: {total_attributes}")
+    print(f"总耗时: {elapsed_time:.1f} 秒")
+    print()
+    
+    if removed_domains:
+        print("删除的 domains:")
+        for domain in removed_domains[:20]:  # 最多显示20个
+            print(f"   - {domain}")
+        if len(removed_domains) > 20:
+            print(f"   ... 还有 {len(removed_domains) - 20} 个")
+        print()
+    
+    print(f"💾 结果已保存到: {output_file}")
+    print()
+    print("=" * 80)
+    print("✨ 处理完成！")
+    print("=" * 80)
+
+
+if __name__ == '__main__':
+    main()

icon_generation/backup/filter_accepted_log.txt

@@ -0,0 +1,383 @@
+================================================================================
+Domain-Attribute 过滤和精简工具
+================================================================================
+
+📖 读取文件: accepted_domains_attributes.txt
+✅ 成功读取 338 个 domains
+
+================================================================================
+🚀 开始处理 (10线程并发)
+   总 domains: 338
+   批次数: 34
+   每批: 10 个 domains
+================================================================================
+
+
+================================================================================
+🔄 处理批次 1/34 (2.9%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 2/34 (5.9%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 3/34 (8.8%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 4/34 (11.8%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 5/34 (14.7%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 6/34 (17.6%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 7/34 (20.6%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.1秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 8/34 (23.5%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.1秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 9/34 (26.5%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.1秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+
+================================================================================
+🔄 处理批次 10/34 (29.4%)
+⏱️  已用时间: 0.0秒 | 预计剩余: 0.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 11/34 (32.4%)
+⏱️  已用时间: 10.9秒 | 预计剩余: 22.8秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 12/34 (35.3%)
+⏱️  已用时间: 11.2秒 | 预计剩余: 20.6秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Landmark Name
+
+================================================================================
+🔄 处理批次 13/34 (38.2%)
+⏱️  已用时间: 11.4秒 | 预计剩余: 18.4秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 14/34 (41.2%)
+⏱️  已用时间: 11.7秒 | 预计剩余: 16.7秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 15/34 (44.1%)
+⏱️  已用时间: 12.2秒 | 预计剩余: 15.4秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 16/34 (47.1%)
+⏱️  已用时间: 13.9秒 | 预计剩余: 15.6秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 17/34 (50.0%)
+⏱️  已用时间: 14.1秒 | 预计剩余: 14.1秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 18/34 (52.9%)
+⏱️  已用时间: 14.6秒 | 预计剩余: 13.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 19/34 (55.9%)
+⏱️  已用时间: 18.8秒 | 预计剩余: 14.9秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 20/34 (58.8%)
+⏱️  已用时间: 19.0秒 | 预计剩余: 13.3秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Citrus Variety
+
+================================================================================
+🔄 处理批次 21/34 (61.8%)
+⏱️  已用时间: 23.1秒 | 预计剩余: 14.3秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Tourist Attraction
+
+================================================================================
+🔄 处理批次 22/34 (64.7%)
+⏱️  已用时间: 25.6秒 | 预计剩余: 14.0秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 23/34 (67.6%)
+⏱️  已用时间: 26.5秒 | 预计剩余: 12.7秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 24/34 (70.6%)
+⏱️  已用时间: 29.2秒 | 预计剩余: 12.2秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Primate Species
+
+================================================================================
+🔄 处理批次 25/34 (73.5%)
+⏱️  已用时间: 29.4秒 | 预计剩余: 10.6秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 8 个
+      删除: 2 个
+      删除的 domains: Satellite Mission Type, Fantasy Creature or Race
+
+================================================================================
+🔄 处理批次 26/34 (76.5%)
+⏱️  已用时间: 30.0秒 | 预计剩余: 9.2秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 27/34 (79.4%)
+⏱️  已用时间: 30.0秒 | 预计剩余: 7.8秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Notable Sacred Sites
+
+================================================================================
+🔄 处理批次 28/34 (82.4%)
+⏱️  已用时间: 30.8秒 | 预计剩余: 6.6秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 29/34 (85.3%)
+⏱️  已用时间: 33.5秒 | 预计剩余: 5.8秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 30/34 (88.2%)
+⏱️  已用时间: 34.2秒 | 预计剩余: 4.6秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 8 个
+      删除: 2 个
+      删除的 domains: Planetary Rover Name, Commercial Aircraft Model
+
+================================================================================
+🔄 处理批次 31/34 (91.2%)
+⏱️  已用时间: 37.1秒 | 预计剩余: 3.6秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 7 个
+      删除: 3 个
+      删除的 domains: Common Amphibian Species (common names), Mammal Species Name, Artwork Title
+
+================================================================================
+🔄 处理批次 32/34 (94.1%)
+⏱️  已用时间: 38.7秒 | 预计剩余: 2.4秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 8 个
+      删除: 2 个
+      删除的 domains: Passenger Persona (Travel Behavior), Solar System Planet
+
+================================================================================
+🔄 处理批次 33/34 (97.1%)
+⏱️  已用时间: 38.7秒 | 预计剩余: 1.2秒
+   当前批次: 10 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+
+================================================================================
+🔄 处理批次 34/34 (100.0%)
+⏱️  已用时间: 39.2秒 | 预计剩余: 0.0秒
+   当前批次: 8 个 domains
+   🤖 调用 LLM 进行过滤和精简...
+   ✅ 处理完成!
+      保留: 6 个
+      删除: 4 个
+      删除的 domains: Jewelry-making Technique, Cultural Attire, Chinese New Year Foods, Maize (Corn) Cultivar/Variety
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Ranching Practices and Systems
+   ✅ 处理完成!
+      保留: 8 个
+      删除: 2 个
+      删除的 domains: Astronomical Observatory Name, Historic Site Name
+   ✅ 处理完成!
+      保留: 7 个
+      删除: 3 个
+      删除的 domains: Role-Playing Game System, Product Variant (Production Method), Cultural Art Traditions
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Lemur Taxon Name (Genus and Species)
+   ✅ 处理完成!
+      保留: 10 个
+      删除: 0 个
+   ✅ 处理完成!
+      保留: 8 个
+      删除: 2 个
+      删除的 domains: Comic Book Series, Power Plant
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Seabird Species
+   ✅ 处理完成!
+      保留: 8 个
+      删除: 0 个
+   ✅ 处理完成!
+      保留: 9 个
+      删除: 1 个
+      删除的 domains: Crop Pest
+
+================================================================================
+📊 合并结果...
+================================================================================
+
+================================================================================
+💾 保存结果...
+================================================================================
+
+================================================================================
+📊 统计信息:
+================================================================================
+原始 domains 数: 338
+保留 domains 数: 309
+删除 domains 数: 29
+总 attributes 数: 4370
+总耗时: 50.7 秒
+
+删除的 domains:
+   - Landmark Name
+   - Citrus Variety
+   - Tourist Attraction
+   - Primate Species
+   - Satellite Mission Type
+   - Fantasy Creature or Race
+   - Notable Sacred Sites
+   - Planetary Rover Name
+   - Commercial Aircraft Model
+   - Common Amphibian Species (common names)
+   - Mammal Species Name
+   - Artwork Title
+   - Passenger Persona (Travel Behavior)
+   - Solar System Planet
+   - Jewelry-making Technique
+   - Cultural Attire
+   - Chinese New Year Foods
+   - Maize (Corn) Cultivar/Variety
+   - Ranching Practices and Systems
+   - Astronomical Observatory Name
+   ... 还有 9 个
+
+💾 结果已保存到: accepted_domains_attributes_filtered.txt
+
+================================================================================
+✨ 处理完成！
+================================================================================

icon_generation/backup/filter_pairs.py

@@ -0,0 +1,402 @@
+#!/usr/bin/env python3
+"""
+使用 LLM 过滤 domain_value_pairs.txt
+移除不对应、抽象概念或标识符类型的 values
+每次处理 100 个 pairs
+"""
+
+import json
+import requests
+from typing import List, Dict, Tuple, Optional
+import time
+import os
+
+
+class PairFilter:
+    """使用 LLM 过滤 domain-value pairs"""
+    
+    def __init__(self, api_key=None, base_url=None, model=None):
+        """初始化 LLM analyzer"""
+        self.api_key = api_key or os.getenv("OPENAI_API_KEY") or os.getenv("AIHUBMIX_API_KEY", "")
+        self.base_url = base_url or os.getenv("OPENAI_BASE_URL", "https://aihubmix.com/v1")
+        self.model = model or os.getenv("OPENAI_MODEL", "gemini-2.5-flash")
+    
+    def filter_pairs_batch(self, pairs: List[Tuple[str, str, int]]) -> Dict:
+        """
+        使用 LLM 批量过滤 pairs
+        
+        Args:
+            pairs: [(domain, value, count), ...]
+        
+        Returns:
+            {
+                'keep': [(domain, value, count), ...],
+                'remove': [(domain, value, count, reason), ...],
+                'reasoning': 'overall explanation'
+            }
+        """
+        prompt = self._build_filter_prompt(pairs)
+        
+        try:
+            response = self._query_llm(prompt)
+            
+            if response:
+                # 清理可能的 markdown 代码块
+                cleaned_response = response.strip()
+                if cleaned_response.startswith('```'):
+                    lines = cleaned_response.split('\n')
+                    cleaned_response = '\n'.join(lines[1:-1] if lines[-1].strip() == '```' else lines[1:])
+                    cleaned_response = cleaned_response.replace('```json', '').replace('```', '').strip()
+                
+                result = json.loads(cleaned_response)
+                
+                # 验证返回格式
+                if 'keep_indices' in result and 'remove_indices' in result:
+                    keep_indices = set(result['keep_indices'])
+                    remove_info = {item['index']: item['reason'] for item in result['remove_indices']}
+                    
+                    keep = []
+                    remove = []
+                    
+                    for idx, (domain, value, count) in enumerate(pairs):
+                        if idx in keep_indices:
+                            keep.append((domain, value, count))
+                        elif idx in remove_info:
+                            remove.append((domain, value, count, remove_info[idx]))
+                        else:
+                            # 如果 LLM 没有明确说明，默认保留
+                            keep.append((domain, value, count))
+                    
+                    return {
+                        'keep': keep,
+                        'remove': remove,
+                        'reasoning': result.get('reasoning', '')
+                    }
+                else:
+                    print(f"      ⚠️  LLM 响应缺少必需字段")
+                    # 默认全部保留
+                    return {
+                        'keep': pairs,
+                        'remove': [],
+                        'reasoning': 'LLM response format error, kept all'
+                    }
+                    
+            else:
+                print(f"      ⚠️  LLM API 调用失败")
+                return {
+                    'keep': pairs,
+                    'remove': [],
+                    'reasoning': 'LLM API failed, kept all'
+                }
+                
+        except json.JSONDecodeError as e:
+            print(f"      ⚠️  LLM 响应不是有效的 JSON: {e}")
+            print(f"      响应: {response[:300]}...")
+            return {
+                'keep': pairs,
+                'remove': [],
+                'reasoning': 'JSON decode error, kept all'
+            }
+        except Exception as e:
+            print(f"      ⚠️  过滤错误: {e}")
+            return {
+                'keep': pairs,
+                'remove': [],
+                'reasoning': f'Error: {str(e)}, kept all'
+            }
+    
+    def _build_filter_prompt(self, pairs: List[Tuple[str, str, int]]) -> str:
+        """构建过滤的 prompt"""
+        
+        # 构建 pairs 列表字符串
+        pairs_str = '\n'.join([
+            f"  {idx}. Domain: \"{domain}\", Value: \"{value}\", Count: {count}"
+            for idx, (domain, value, count) in enumerate(pairs)
+        ])
+        
+        prompt = f"""You are a data quality expert. Given a list of domain-value pairs, identify which values should be REMOVED because they:
+
+1. **Don't match the domain**: The value doesn't truly belong to or represent the stated domain
+2. **Are abstract concepts**: Generic, vague, or non-specific values (e.g., "Unknown", "Other", "Various")
+3. **Are mere identifiers**: Values that are just labels, codes, or IDs without semantic meaning (e.g., "TruckA", "TruckB", "Final Cost", "Option1", "Item #123")
+
+Domain-Value Pairs to analyze:
+{pairs_str}
+
+Please analyze each pair and return ONLY the indices that should be:
+- **KEPT**: Concrete, specific, meaningful values that clearly belong to their domain
+- **REMOVED**: Values that fall into any of the three categories above
+
+Return your response in the following JSON format ONLY (no markdown, no extra text):
+{{
+    "keep_indices": [0, 2, 5, ...],
+    "remove_indices": [
+        {{"index": 1, "reason": "abstract concept"}},
+        {{"index": 3, "reason": "mere identifier"}},
+        {{"index": 4, "reason": "doesn't match domain"}},
+        ...
+    ],
+    "reasoning": "Brief summary of filtering approach"
+}}
+
+Examples to guide your decision:
+- KEEP: "US State: California", "Sport Type: Basketball", "Industry Sector: Healthcare"
+- REMOVE: "US State: Unknown" (abstract), "Vehicle: TruckA" (identifier), "Country: New York" (doesn't match - it's a city)
+- REMOVE: "Category: Other", "Type: Various", "Name: Item1", "Cost: Final Cost" (all abstract/identifiers)
+"""
+        
+        return prompt
+    
+    def _query_llm(self, prompt: str) -> Optional[str]:
+        """查询 LLM API"""
+        headers = {
+            'Authorization': f'Bearer {self.api_key}',
+            'Content-Type': 'application/json'
+        }
+        
+        data = {
+            'model': self.model,
+            'messages': [
+                {
+                    'role': 'system', 
+                    'content': 'You are a data quality expert specialized in filtering and validating domain-value pairs. Always return valid JSON format only, without any markdown formatting or extra text.'
+                },
+                {
+                    'role': 'user', 
+                    'content': prompt
+                }
+            ],
+            'temperature': 0.2
+        }
+        
+        try:
+            response = requests.post(
+                f'{self.base_url}/chat/completions',
+                headers=headers,
+                json=data,
+                timeout=60
+            )
+            response.raise_for_status()
+            
+            result = response.json()
+            return result['choices'][0]['message']['content'].strip()
+            
+        except requests.exceptions.Timeout:
+            print("      ❌ LLM API 超时")
+            return None
+        except requests.exceptions.HTTPError as e:
+            print(f"      ❌ LLM API HTTP 错误: {e}")
+            return None
+        except requests.exceptions.RequestException as e:
+            print(f"      ❌ LLM API 请求错误: {e}")
+            return None
+        except KeyError as e:
+            print(f"      ❌ LLM API 响应格式错误: {e}")
+            return None
+
+
+def load_pairs(file_path: str) -> List[Tuple[str, str, int]]:
+    """加载 domain_value_pairs.txt"""
+    pairs = []
+    
+    with open(file_path, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                continue
+            
+            # 解析 CSV 格式，处理带引号的字段
+            parts = []
+            current = []
+            in_quotes = False
+            
+            for char in line:
+                if char == '"':
+                    in_quotes = not in_quotes
+                elif char == ',' and not in_quotes:
+                    parts.append(''.join(current))
+                    current = []
+                else:
+                    current.append(char)
+            parts.append(''.join(current))
+            
+            if len(parts) >= 3:
+                domain = parts[0].strip()
+                value = parts[1].strip()
+                count = int(parts[2].strip())
+                pairs.append((domain, value, count))
+    
+    return pairs
+
+
+def save_filtered_pairs(pairs: List[Tuple[str, str, int]], output_file: str):
+    """保存过滤后的 pairs"""
+    with open(output_file, 'w', encoding='utf-8') as f:
+        for domain, value, count in pairs:
+            # 处理可能包含逗号的字段
+            if ',' in value:
+                value = f'"{value}"'
+            if ',' in domain:
+                domain = f'"{domain}"'
+            f.write(f"{domain},{value},{count}\n")
+
+
+def main():
+    print("=" * 80)
+    print("Domain-Value Pairs 过滤器")
+    print("=" * 80)
+    print()
+    
+    input_file = 'domain_value_pairs.txt'
+    output_file = 'domain_value_pairs_filtered.txt'
+    temp_file = 'domain_value_pairs_filtered_temp.txt'
+    removed_file = 'domain_value_pairs_removed.txt'
+    
+    # 加载数据
+    print(f"📖 读取文件: {input_file}")
+    pairs = load_pairs(input_file)
+    print(f"✅ 成功读取 {len(pairs)} 个 pairs")
+    print()
+    
+    # 检查是否有临时文件（断点续传）
+    filtered_pairs = []
+    removed_pairs = []
+    start_idx = 0
+    
+    if os.path.exists(temp_file):
+        print(f"📂 发现临时文件，尝试恢复进度...")
+        try:
+            filtered_pairs = load_pairs(temp_file)
+            start_idx = len(filtered_pairs)
+            print(f"✅ 已恢复 {start_idx} 个 pairs 的处理结果")
+        except Exception as e:
+            print(f"⚠️  临时文件读取失败: {e}，从头开始")
+            filtered_pairs = []
+            start_idx = 0
+    
+    if os.path.exists(removed_file):
+        try:
+            with open(removed_file, 'r', encoding='utf-8') as f:
+                for line in f:
+                    if line.strip():
+                        removed_pairs.append(line.strip())
+        except:
+            pass
+    
+    # 初始化过滤器
+    filter_obj = PairFilter()
+    
+    # 批量处理
+    batch_size = 100
+    total_batches = (len(pairs) - start_idx + batch_size - 1) // batch_size
+    
+    print("=" * 80)
+    print(f"🚀 开始过滤处理")
+    print(f"   总 pairs 数: {len(pairs)}")
+    print(f"   已处理: {start_idx}")
+    print(f"   待处理: {len(pairs) - start_idx}")
+    print(f"   批次大小: {batch_size}")
+    print(f"   总批次数: {total_batches}")
+    print("=" * 80)
+    print()
+    
+    start_time = time.time()
+    
+    for batch_idx in range(0, len(pairs) - start_idx, batch_size):
+        actual_idx = start_idx + batch_idx
+        batch = pairs[actual_idx:actual_idx + batch_size]
+        current_batch_num = batch_idx // batch_size + 1
+        
+        # 进度信息
+        progress = (actual_idx + len(batch)) / len(pairs) * 100
+        elapsed = time.time() - start_time
+        avg_time = elapsed / (batch_idx + batch_size) if batch_idx > 0 else 0
+        remaining = avg_time * (len(pairs) - start_idx - batch_idx - len(batch))
+        
+        print(f"{'=' * 80}")
+        print(f"🔄 批次 {current_batch_num}/{total_batches}")
+        print(f"   进度: {actual_idx + len(batch)}/{len(pairs)} ({progress:.1f}%)")
+        print(f"   已用时间: {elapsed:.1f}秒 | 预计剩余: {remaining:.1f}秒")
+        print(f"   当前批次: {len(batch)} 个 pairs")
+        
+        # 显示前 3 个示例
+        print(f"   示例:")
+        for i, (domain, value, count) in enumerate(batch[:3]):
+            print(f"      {i+1}. {domain}: {value} (count: {count})")
+        
+        print(f"   🤖 调用 LLM 进行过滤...")
+        
+        # 调用 LLM 过滤
+        result = filter_obj.filter_pairs_batch(batch)
+        
+        keep_count = len(result['keep'])
+        remove_count = len(result['remove'])
+        
+        print(f"   ✅ 过滤完成!")
+        print(f"      保留: {keep_count} 个")
+        print(f"      移除: {remove_count} 个")
+        
+        # 显示移除的示例
+        if result['remove']:
+            print(f"      移除示例:")
+            for domain, value, count, reason in result['remove'][:3]:
+                print(f"         - {domain}: {value} ({reason})")
+        
+        # 更新结果
+        filtered_pairs.extend(result['keep'])
+        
+        # 保存移除的记录
+        if result['remove']:
+            with open(removed_file, 'a', encoding='utf-8') as f:
+                for domain, value, count, reason in result['remove']:
+                    if ',' in value:
+                        value = f'"{value}"'
+                    if ',' in domain:
+                        domain = f'"{domain}"'
+                    f.write(f"{domain},{value},{count},{reason}\n")
+            removed_pairs.extend(result['remove'])
+        
+        # 保存临时文件
+        save_filtered_pairs(filtered_pairs, temp_file)
+        print(f"   💾 已保存临时结果")
+    
+    # 保存最终结果
+    print(f"\n{'=' * 80}")
+    print("💾 保存最终结果...")
+    save_filtered_pairs(filtered_pairs, output_file)
+    
+    # 删除临时文件
+    if os.path.exists(temp_file):
+        os.remove(temp_file)
+    
+    elapsed_time = time.time() - start_time
+    
+    # 统计信息
+    print(f"\n{'=' * 80}")
+    print("📊 过滤完成！统计信息：")
+    print("=" * 80)
+    print(f"✅ 保留 pairs: {len(filtered_pairs)} 个")
+    print(f"❌ 移除 pairs: {len(removed_pairs)} 个")
+    print(f"📋 原始 pairs: {len(pairs)} 个")
+    print(f"📉 过滤比例: {len(removed_pairs)/len(pairs)*100:.1f}%")
+    print(f"⏱️  总耗时: {elapsed_time:.1f} 秒")
+    
+    print(f"\n💾 文件保存:")
+    print(f"   保留的 pairs: {output_file}")
+    print(f"   移除的 pairs: {removed_file}")
+    
+    # 显示前 10 个保留的
+    print(f"\n🏆 Top 10 保留的 pairs:")
+    for i, (domain, value, count) in enumerate(filtered_pairs[:10], 1):
+        print(f"   {i:2d}. {domain}: {value} ({count:,})")
+    
+    print(f"\n{'=' * 80}")
+    print("✨ 过滤完成！")
+    print("=" * 80)
+
+
+if __name__ == '__main__':
+    main()
+
+
+

icon_generation/backup/image_batch_generator.py

@@ -0,0 +1,571 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+图像批量生成Pipeline
+根据topic_style.json配置，批量生成适合infographic装饰的图像
+"""
+
+import os
+import json
+import random
+import sys
+import time
+from typing import Dict, List, Tuple
+from openai import OpenAI
+from concurrent.futures import ThreadPoolExecutor
+from google import genai
+from google.genai import types
+from PIL import Image, ImageDraw
+from io import BytesIO
+import numpy as np
+from collections import Counter
+
+# 添加项目根目录到路径
+# sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+# from config import api_key, base_url
+
+api_key = 'xxx'
+base_url = "https://aihubmix.com/v1"
+
+class ImageBatchGenerator:
+    def __init__(self):
+        """初始化生成器"""
+        # OpenAI client for text generation
+        self.openai_client = OpenAI(
+            api_key=api_key,
+            base_url=base_url,
+        )
+        
+        # Gemini client for image generation
+        self.genai_client = genai.Client(
+            api_key=api_key,
+            http_options={"base_url": "https://aihubmix.com/gemini"},
+        )
+        
+        # 加载topic_style配置
+        self.config_path = os.path.join(
+            os.path.dirname(os.path.dirname(os.path.abspath(__file__))),
+            'generator', 'topic_style.json'
+        )
+        self.load_config()
+        
+        # 输出目录
+        self.output_dir = os.path.join(
+            os.path.dirname(os.path.dirname(os.path.abspath(__file__))),
+            'gen_output'
+        )
+        os.makedirs(self.output_dir, exist_ok=True)
+        
+        # 设计prompt模板
+        self.design_prompt_template = """
+[TASK START]
+OBJECTIVE: Generate a text-to-image prompt for a single, isolated clipart icon based on the provided inputs.
+
+INPUTS:
+Topic: {topic}
+Style Keyword: {style_keyword}
+Concept: {concept}
+
+PROCESS:
+Write a text-to-image prompt describing this concept, rendered using the specified Style Keyword.
+
+CONSTRAINTS:
+- The output must be a single icon or a small, unified group of objects
+- The icon MUST be isolated on a pure white background (#FFFFFF)
+- No shadows, textures or patterns in the background
+- The background must be completely clean and empty
+- The final prompt must be concise and descriptive
+
+REQUIRED OUTPUT:
+[The final text-to-image prompt, make sure to specify "on pure white background" in the prompt]
+
+[TASK END]
+"""
+        
+        # 概念生成prompt
+        self.concept_generation_prompt = """
+Generate 10 different concrete concepts for the topic "{topic}".
+
+Requirements:
+1. Each concept must be a specific, tangible object or clear visual scene
+2. Use detailed descriptions (e.g. "stethoscope on medical chart" vs "medical")
+3. Focus on real-world items, tools, places or situations
+4. Each concept should be immediately recognizable and relatable
+5. Concepts should work well as simple icons or decorative elements
+6. Keep descriptions concise but specific
+
+Return in this format:
+1. [concept1]
+2. [concept2]
+3. [concept3]
+...
+10. [concept10]
+"""
+        
+        # 设计评判prompt
+        self.design_evaluation_prompt = """
+Evaluate the following design concepts and select the 5 best ones for infographic decoration.
+
+Evaluation criteria:
+1. Visual clarity: Easy to recognize and understand
+2. Decorative value: Suitable as decorative elements without interfering with main information
+3. Universality: Broad applicability
+
+Design concept list:
+{concepts}
+
+Select the 5 best concepts and return in this format:
+Selected concepts:
+1. [concept name]
+2. [concept name]
+3. [concept name]
+4. [concept name]
+5. [concept name]
+"""
+
+    def load_config(self):
+        """加载topic_style配置文件"""
+        with open(self.config_path, 'r', encoding='utf-8') as f:
+            self.config = json.load(f)
+        print(f"✅ 加载配置: {len(self.config)} 个风格类别")
+
+    def select_random_category_and_elements(self) -> Tuple[str, str, str]:
+        """随机选择category、keyword和topic"""
+        category = random.choice(list(self.config.keys()))
+        category_data = self.config[category]
+        keyword = random.choice(category_data['keywords'])
+        topic = random.choice(category_data['topics'])
+        
+        print(f"🎯 选中: {category} | {keyword} | {topic}")
+        return category, keyword, topic
+
+    def generate_concepts(self, topic: str) -> List[str]:
+        """使用ChatGPT生成10个概念"""
+        print(f"🧠 生成概念...")
+        
+        response = self.openai_client.chat.completions.create(
+            model="gpt-5-mini",
+            messages=[
+                {"role": "user", "content": self.concept_generation_prompt.format(topic=topic)}
+            ],
+            temperature=0.8
+        )
+        
+        content = response.choices[0].message.content
+        
+        # 解析概念列表 - 修复方括号解析问题
+        concepts = []
+        lines = content.strip().split('\n')
+        for line in lines:
+            line = line.strip()
+            if line and (line[0].isdigit() or line.startswith('-')):
+                # 提取方括号内的内容
+                if '[' in line and ']' in line:
+                    start = line.find('[')
+                    end = line.find(']')
+                    if start != -1 and end != -1 and end > start:
+                        concept = line[start+1:end].strip()
+                        if concept:
+                            concepts.append(concept)
+                else:
+                    # 如果没有方括号，提取序号后的内容
+                    concept = line.split('.', 1)[-1].strip()
+                    if concept:
+                        concepts.append(concept)
+        
+        print(f"✅ 生成 {len(concepts)} 个概念")
+        return concepts[:10]
+
+    def evaluate_and_select_concepts(self, concepts: List[str]) -> List[str]:
+        """评判并选择5个最佳概念"""
+        print(f"🔍 评判概念...")
+        
+        concepts_text = ""
+        for i, concept in enumerate(concepts, 1):
+            concepts_text += f"{i}. {concept}\n"
+        
+        response = self.openai_client.chat.completions.create(
+            model="gpt-5-mini",
+            messages=[
+                {"role": "user", "content": self.design_evaluation_prompt.format(concepts=concepts_text)}
+            ],
+            temperature=0.3
+        )
+        
+        content = response.choices[0].message.content
+        
+        # 解析选中的概念
+        selected_concepts = []
+        lines = content.strip().split('\n')
+        
+        for line in lines:
+            line = line.strip()
+            if line and line[0].isdigit() and '.' in line:
+                concept_name = line.split('.', 1)[1].strip()
+                # 在原始概念中查找匹配
+                for concept in concepts:
+                    if concept_name.lower() in concept.lower() or concept.lower() in concept_name.lower():
+                        if concept not in selected_concepts:
+                            selected_concepts.append(concept)
+                            break
+        
+        # 如果解析不足5个，随机补充
+        if len(selected_concepts) < 5:
+            remaining = [c for c in concepts if c not in selected_concepts]
+            selected_concepts.extend(random.sample(remaining, min(5 - len(selected_concepts), len(remaining))))
+        
+        print(f"✅ 选中 {len(selected_concepts[:5])} 个概念")
+        return selected_concepts[:5]
+
+    def detect_background_color(self, image: Image.Image) -> tuple:
+        """检测图像的背景颜色，返回(背景色, 是否为杂乱背景)"""
+        # 获取图像尺寸
+        width, height = image.size
+        
+        # 采样边界点
+        sample_points = []
+        
+        # 四个角
+        sample_points.extend([
+            (0, 0), (width-1, 0), (0, height-1), (width-1, height-1)
+        ])
+        
+        # 边界中点
+        sample_points.extend([
+            (width//2, 0), (width//2, height-1),  # 上下边中点
+            (0, height//2), (width-1, height//2)  # 左右边中点
+        ])
+        
+        # 边界线采样（每边采样10个点）
+        for i in range(1, 10):
+            ratio = i / 10.0
+            # 上边
+            sample_points.append((int(width * ratio), 0))
+            # 下边
+            sample_points.append((int(width * ratio), height-1))
+            # 左边
+            sample_points.append((0, int(height * ratio)))
+            # 右边
+            sample_points.append((width-1, int(height * ratio)))
+        
+        # 获取所有采样点的颜色
+        colors = []
+        for x, y in sample_points:
+            if 0 <= x < width and 0 <= y < height:
+                pixel = image.getpixel((x, y))
+                if isinstance(pixel, int):  # 灰度图
+                    colors.append((pixel, pixel, pixel))
+                elif len(pixel) >= 3:  # RGB或RGBA
+                    colors.append(pixel[:3])
+        
+        # 统计颜色众数
+        color_counts = Counter(colors)
+        if color_counts:
+            most_common_color, most_common_count = color_counts.most_common(1)[0]
+            total_samples = len(colors)
+            
+            # 计算众数颜色占比
+            ratio = most_common_count / total_samples
+            
+            # 如果众数颜色占比小于50%，认为背景杂乱
+            is_messy = ratio < 0.5
+            
+            return most_common_color, is_messy
+        
+        # 默认返回白色，非杂乱
+        return (255, 255, 255), False
+
+    def optimized_flood_fill_remove_background(self, image: Image.Image, bg_color: tuple, tolerance: int = 30) -> Image.Image:
+        """使用优化的flood fill算法从边界去除背景色"""
+        # 转换为RGBA模式
+        if image.mode != 'RGBA':
+            image = image.convert('RGBA')
+        
+        # 转换为numpy数组
+        data = np.array(image, dtype=np.uint8)
+        height, width = data.shape[:2]
+        
+        # 创建访问标记数组
+        visited = np.zeros((height, width), dtype=bool)
+        
+        # 预计算颜色距离的平方（避免开方运算）
+        def color_distance_squared(c1, c2):
+            """计算颜色距离的平方，避免开方运算提高性能"""
+            return sum((int(a) - int(b)) ** 2 for a, b in zip(c1[:3], c2[:3]))
+        
+        tolerance_squared = tolerance * tolerance
+        
+        def is_background_color(pixel_color):
+            """判断是否为背景色，使用平方距离比较"""
+            return color_distance_squared(pixel_color[:3], bg_color) <= tolerance_squared
+        
+        def optimized_flood_fill(start_x, start_y):
+            """优化的flood fill算法，使用栈而非递归，批量处理"""
+            if (start_y >= height or start_x >= width or 
+                start_y < 0 or start_x < 0 or 
+                visited[start_y, start_x]):
+                return
+            
+            # 使用deque作为栈，性能更好
+            from collections import deque
+            stack = deque([(start_x, start_y)])
+            pixels_to_clear = []
+            
+            while stack:
+                x, y = stack.pop()
+                
+                # 边界检查
+                if x < 0 or x >= width or y < 0 or y >= height or visited[y, x]:
+                    continue
+                
+                current_color = data[y, x]
+                
+                # 检查颜色是否在容差范围内
+                if not is_background_color(current_color):
+                    continue
+                
+                # 标记为已访问
+                visited[y, x] = True
+                pixels_to_clear.append((x, y))
+                
+                # 添加相邻像素到栈中（4连通）
+                stack.extend([
+                    (x+1, y), (x-1, y), (x, y+1), (x, y-1)
+                ])
+            
+            # 批量设置像素为透明
+            for x, y in pixels_to_clear:
+                data[y, x] = (0, 0, 0, 0)
+        
+        print(f"    🌊 优化Flood Fill处理...")
+        
+        # 从边界开始flood fill，优化边界遍历
+        # 上边和下边
+        for x in range(0, width, 2):  # 每隔一个像素采样，提高性能
+            optimized_flood_fill(x, 0)
+            optimized_flood_fill(x, height-1)
+        
+        # 左边和右边
+        for y in range(0, height, 2):  # 每隔一个像素采样，提高性能
+            optimized_flood_fill(0, y)
+            optimized_flood_fill(width-1, y)
+        
+        # 补充处理边界的奇数位置
+        for x in range(1, width, 2):
+            if not visited[0, x]:
+                optimized_flood_fill(x, 0)
+            if not visited[height-1, x]:
+                optimized_flood_fill(x, height-1)
+        
+        for y in range(1, height, 2):
+            if not visited[y, 0]:
+                optimized_flood_fill(0, y)
+            if not visited[y, width-1]:
+                optimized_flood_fill(width-1, y)
+        
+        # 转换回PIL图像
+        return Image.fromarray(data, 'RGBA')
+
+    def crop_transparent_borders(self, image: Image.Image) -> Image.Image:
+        """裁剪透明边界，去除多余区域"""
+        if image.mode != 'RGBA':
+            return image
+        
+        # 转换为numpy数组
+        data = np.array(image)
+        
+        # 获取alpha通道
+        alpha = data[:, :, 3]
+        
+        # 找到非透明像素的边界
+        non_transparent = np.where(alpha > 0)
+        
+        if len(non_transparent[0]) == 0:
+            # 如果图像完全透明，返回最小尺寸
+            return image.crop((0, 0, 1, 1))
+        
+        # 计算边界框
+        min_y, max_y = non_transparent[0].min(), non_transparent[0].max()
+        min_x, max_x = non_transparent[1].min(), non_transparent[1].max()
+        
+        # 添加小的边距（5像素）
+        padding = 5
+        width, height = image.size
+        
+        min_x = max(0, min_x - padding)
+        min_y = max(0, min_y - padding)
+        max_x = min(width - 1, max_x + padding)
+        max_y = min(height - 1, max_y + padding)
+        
+        # 裁剪图像
+        cropped = image.crop((min_x, min_y, max_x + 1, max_y + 1))
+        
+        return cropped
+
+    def post_process_image(self, image: Image.Image) -> Image.Image:
+        """后处理图像：去除背景并裁剪多余区域，如果背景杂乱则返回None"""
+        print(f"    🔧 后处理图像...")
+        
+        # 检测背景颜色和杂乱程度
+        bg_color, is_messy = self.detect_background_color(image)
+        
+        if is_messy:
+            print(f"    ❌ 检测到杂乱背景，抛弃此图片")
+            return None
+        
+        print(f"    📊 检测到背景色: {bg_color}")
+        
+        # 使用优���的flood fill去除背景
+        processed_image = self.optimized_flood_fill_remove_background(image, bg_color, tolerance=30)
+        
+        # 裁剪透明边界
+        cropped_image = self.crop_transparent_borders(processed_image)
+        
+        original_size = image.size
+        final_size = cropped_image.size
+        print(f"    ✂️ 尺寸调整: {original_size} → {final_size}")
+        
+        return cropped_image
+
+    def generate_prompt_and_image(self, concept: str, topic: str, keyword: str, category: str) -> str:
+        """为单个概念生成prompt并生成图像"""
+        print(f"  🎨 处理: {concept[:50]}...")
+        
+        # 生成设计prompt
+        prompt = self.design_prompt_template.format(
+            topic=topic,
+            style_keyword=keyword,
+            concept=concept
+        )
+        
+        response = self.openai_client.chat.completions.create(
+            model="gpt-5-mini",
+            messages=[
+                {"role": "user", "content": prompt}
+            ],
+            temperature=0.7
+        )
+        
+        image_prompt = response.choices[0].message.content.strip()
+        
+        # 生成图像使用imagen-4.0，带重试机制
+        max_retries = 5
+        retry_delay = 5  # 秒
+        response = None
+        
+        for attempt in range(max_retries):
+            try:
+                print(f"    🖼️ 生成图像 (尝试 {attempt + 1}/{max_retries})...")
+                response = self.genai_client.models.generate_images(
+                    model='imagen-4.0-fast-generate-001',
+                    prompt=image_prompt,
+                    config=types.GenerateImagesConfig(
+                        number_of_images=1,
+                        aspect_ratio="1:1",
+                    )
+                )
+                # 如果成功，跳出重试循环
+                if response and hasattr(response, 'generated_images') and response.generated_images:
+                    print(f"    ✅ 图像生成成功")
+                    break
+                else:
+                    print(f"    ⚠️ 图像生成返回空结果")
+                    if attempt < max_retries - 1:
+                        print(f"    ⏳ 等待 {retry_delay} 秒后重试...")
+                        time.sleep(retry_delay)
+                    
+            except Exception as e:
+                print(f"    ❌ 图像生成失败 (尝试 {attempt + 1}/{max_retries}): {str(e)}")
+                if attempt < max_retries - 1:
+                    print(f"    ⏳ 等待 {retry_delay} 秒后重试...")
+                    time.sleep(retry_delay)
+                else:
+                    print(f"    💀 所有重试均失败，放弃生成此图像")
+                    return None
+        
+        # 保存图像
+        if response and hasattr(response, 'generated_images') and response.generated_images:
+            generated_image = response.generated_images[0]
+            image = Image.open(BytesIO(generated_image.image.image_bytes))
+            
+            # 后处理图像：去除背景
+            processed_image = self.post_process_image(image)
+            
+            # 如果图像被抛弃（杂乱背景），返回None
+            if processed_image is None:
+                print(f"    🗑️ 图片已抛弃")
+                return None
+            
+            # 构建文件名 - 使用连字符连接，下划线替换空格
+            safe_topic = topic.replace(' ', '_')
+            safe_category = category.replace(' ', '_')
+            safe_concept = concept[:30].replace(' ', '_')
+            
+            # 移除非字母数字和允许的字符
+            safe_topic = "".join(c for c in safe_topic if c.isalnum() or c in ('_', '-')).strip('_-')
+            safe_category = "".join(c for c in safe_category if c.isalnum() or c in ('_', '-')).strip('_-')
+            safe_concept = "".join(c for c in safe_concept if c.isalnum() or c in ('_', '-')).strip('_-')
+            
+            timestamp = int(time.time())
+            filename = f"{safe_topic}-{safe_category}-{safe_concept}-{timestamp}.png"
+            filepath = os.path.join(self.output_dir, filename)
+            
+            processed_image.save(filepath)
+            print(f"    ✅ 保存: {os.path.basename(filepath)}")
+            return filepath
+        
+        return None
+
+    def run_pipeline(self) -> Dict:
+        """运行完整的批量生成pipeline"""
+        print("🚀 开始图像批量生成Pipeline")
+        
+        # 1. 随机选择category、keyword和topic
+        category, keyword, topic = self.select_random_category_and_elements()
+        
+        # 2. 生成10个概念
+        concepts = self.generate_concepts(topic)
+        
+        # 3. 评判并选择5个最佳概念
+        selected_concepts = self.evaluate_and_select_concepts(concepts)
+        
+        # 4. 并行生成prompt和图像
+        print(f"🖼️ 并行生成 {len(selected_concepts)} 张图像...")
+        generated_files = []
+        
+        with ThreadPoolExecutor(max_workers=3) as executor:
+            futures = []
+            for concept in selected_concepts:
+                future = executor.submit(
+                    self.generate_prompt_and_image, 
+                    concept, topic, keyword, category
+                )
+                futures.append(future)
+            
+            for future in futures:
+                result = future.result()
+                if result:  # 只有成功生成且未被抛弃的图片才会被添加
+                    generated_files.append(result)
+        
+        print(f"✅ 完成! 生成 {len(generated_files)} 张图像")
+        
+        return {
+            'category': category,
+            'keyword': keyword,
+            'topic': topic,
+            'generated_images': len(generated_files),
+            'output_files': generated_files
+        }
+
+
+def main():
+    """主函数"""
+    random.seed(int(time.time()))
+    generator = ImageBatchGenerator()
+    for i in range(1000):
+        result = generator.run_pipeline()
+    print(f"📊 结果: {result['generated_images']} 张图像已保存")
+
+
+if __name__ == "__main__":
+    main()

icon_generation/backup/image_batch_generator_backup.py

@@ -0,0 +1,511 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+图像批量生成Pipeline
+根据topic_style.json配置，批量生成适合infographic装饰的图像
+"""
+
+import os
+import json
+import random
+import sys
+import time
+from typing import Dict, List, Tuple
+from openai import OpenAI
+from concurrent.futures import ThreadPoolExecutor
+from google import genai
+from google.genai import types
+from PIL import Image, ImageDraw
+from io import BytesIO
+import numpy as np
+from collections import Counter
+
+# 添加项目根目录到路径
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+from config import api_key, base_url
+
+class ImageBatchGenerator:
+    def __init__(self):
+        """初始化生成器"""
+        # OpenAI client for text generation
+        self.openai_client = OpenAI(
+            api_key=api_key,
+            base_url=base_url,
+        )
+        
+        # Gemini client for image generation
+        self.genai_client = genai.Client(
+            api_key=api_key,
+            http_options={"base_url": "https://aihubmix.com/gemini"},
+        )
+        
+        # 加载topic_style配置
+        self.config_path = os.path.join(
+            os.path.dirname(os.path.dirname(os.path.abspath(__file__))),
+            'generator', 'topic_style.json'
+        )
+        self.load_config()
+        
+        # 输出目录
+        self.output_dir = os.path.join(
+            os.path.dirname(os.path.dirname(os.path.abspath(__file__))),
+            'gen_output'
+        )
+        os.makedirs(self.output_dir, exist_ok=True)
+        
+        # 设计prompt模板
+        self.design_prompt_template = """
+[TASK START]
+OBJECTIVE: Generate a text-to-image prompt for a single, isolated clipart icon based on the provided inputs.
+
+INPUTS:
+Topic: {topic}
+Style Keyword: {style_keyword}
+Concept: {concept}
+
+PROCESS:
+Write a text-to-image prompt describing this concept, rendered using the specified Style Keyword.
+
+CONSTRAINTS:
+- The output must be a single icon or a small, unified group of objects
+- The icon MUST be isolated on a pure white background (#FFFFFF)
+- No shadows, textures or patterns in the background
+- The background must be completely clean and empty
+- The final prompt must be concise and descriptive
+
+REQUIRED OUTPUT:
+[The final text-to-image prompt, make sure to specify "on pure white background" in the prompt]
+
+[TASK END]
+"""
+        
+        # 概念生成prompt
+        self.concept_generation_prompt = """
+Generate 10 different concrete concepts for the topic "{topic}".
+
+Requirements:
+1. Each concept must be a specific, tangible object or clear visual scene
+2. Use detailed descriptions (e.g. "stethoscope on medical chart" vs "medical")
+3. Focus on real-world items, tools, places or situations
+4. Each concept should be immediately recognizable and relatable
+5. Concepts should work well as simple icons or decorative elements
+6. Keep descriptions concise but specific
+
+Return in this format:
+1. [concept1]
+2. [concept2]
+3. [concept3]
+...
+10. [concept10]
+"""
+        
+        # 设计评判prompt
+        self.design_evaluation_prompt = """
+Evaluate the following design concepts and select the 5 best ones for infographic decoration.
+
+Evaluation criteria:
+1. Visual clarity: Easy to recognize and understand
+2. Decorative value: Suitable as decorative elements without interfering with main information
+3. Universality: Broad applicability
+
+Design concept list:
+{concepts}
+
+Select the 5 best concepts and return in this format:
+Selected concepts:
+1. [concept name]
+2. [concept name]
+3. [concept name]
+4. [concept name]
+5. [concept name]
+"""
+
+    def load_config(self):
+        """加载topic_style配置文件"""
+        with open(self.config_path, 'r', encoding='utf-8') as f:
+            self.config = json.load(f)
+        print(f"✅ 加载配置: {len(self.config)} 个风格类别")
+
+    def select_random_category_and_elements(self) -> Tuple[str, str, str]:
+        """随机选择category、keyword和topic"""
+        category = random.choice(list(self.config.keys()))
+        category_data = self.config[category]
+        keyword = random.choice(category_data['keywords'])
+        topic = random.choice(category_data['topics'])
+        
+        print(f"🎯 选中: {category} | {keyword} | {topic}")
+        return category, keyword, topic
+
+    def generate_concepts(self, topic: str) -> List[str]:
+        """使用ChatGPT生成10个概念"""
+        print(f"🧠 生成概念...")
+        
+        response = self.openai_client.chat.completions.create(
+            model="gpt-5-mini",
+            messages=[
+                {"role": "user", "content": self.concept_generation_prompt.format(topic=topic)}
+            ],
+            temperature=0.8
+        )
+        
+        content = response.choices[0].message.content
+        
+        # 解析概念列表 - 修复方括号解析问题
+        concepts = []
+        lines = content.strip().split('\n')
+        for line in lines:
+            line = line.strip()
+            if line and (line[0].isdigit() or line.startswith('-')):
+                # 提取方括号内的内容
+                if '[' in line and ']' in line:
+                    start = line.find('[')
+                    end = line.find(']')
+                    if start != -1 and end != -1 and end > start:
+                        concept = line[start+1:end].strip()
+                        if concept:
+                            concepts.append(concept)
+                else:
+                    # 如果没有方括号，提取序号后的内容
+                    concept = line.split('.', 1)[-1].strip()
+                    if concept:
+                        concepts.append(concept)
+        
+        print(f"✅ 生成 {len(concepts)} 个概念")
+        return concepts[:10]
+
+    def evaluate_and_select_concepts(self, concepts: List[str]) -> List[str]:
+        """评判并选择5个最佳概念"""
+        print(f"🔍 评判概念...")
+        
+        concepts_text = ""
+        for i, concept in enumerate(concepts, 1):
+            concepts_text += f"{i}. {concept}\n"
+        
+        response = self.openai_client.chat.completions.create(
+            model="gpt-5-mini",
+            messages=[
+                {"role": "user", "content": self.design_evaluation_prompt.format(concepts=concepts_text)}
+            ],
+            temperature=0.3
+        )
+        
+        content = response.choices[0].message.content
+        
+        # 解析选中的概念
+        selected_concepts = []
+        lines = content.strip().split('\n')
+        
+        for line in lines:
+            line = line.strip()
+            if line and line[0].isdigit() and '.' in line:
+                concept_name = line.split('.', 1)[1].strip()
+                # 在原始概念中查找匹配
+                for concept in concepts:
+                    if concept_name.lower() in concept.lower() or concept.lower() in concept_name.lower():
+                        if concept not in selected_concepts:
+                            selected_concepts.append(concept)
+                            break
+        
+        # 如果解析不足5个，随机补充
+        if len(selected_concepts) < 5:
+            remaining = [c for c in concepts if c not in selected_concepts]
+            selected_concepts.extend(random.sample(remaining, min(5 - len(selected_concepts), len(remaining))))
+        
+        print(f"✅ 选中 {len(selected_concepts[:5])} 个概念")
+        return selected_concepts[:5]
+
+    def detect_background_color(self, image: Image.Image) -> tuple:
+        """检测图像的背景颜色，返回(背景色, 是否为杂乱背景)"""
+        # 获取图像尺寸
+        width, height = image.size
+        
+        # 采样边界点
+        sample_points = []
+        
+        # 四个角
+        sample_points.extend([
+            (0, 0), (width-1, 0), (0, height-1), (width-1, height-1)
+        ])
+        
+        # 边界中点
+        sample_points.extend([
+            (width//2, 0), (width//2, height-1),  # 上下边中点
+            (0, height//2), (width-1, height//2)  # 左右边中点
+        ])
+        
+        # 边界线采样（每边采样10个点）
+        for i in range(1, 10):
+            ratio = i / 10.0
+            # 上边
+            sample_points.append((int(width * ratio), 0))
+            # 下边
+            sample_points.append((int(width * ratio), height-1))
+            # 左边
+            sample_points.append((0, int(height * ratio)))
+            # 右边
+            sample_points.append((width-1, int(height * ratio)))
+        
+        # 获取所有采样点的颜色
+        colors = []
+        for x, y in sample_points:
+            if 0 <= x < width and 0 <= y < height:
+                pixel = image.getpixel((x, y))
+                if isinstance(pixel, int):  # 灰度图
+                    colors.append((pixel, pixel, pixel))
+                elif len(pixel) >= 3:  # RGB或RGBA
+                    colors.append(pixel[:3])
+        
+        # 统计颜色众数
+        color_counts = Counter(colors)
+        if color_counts:
+            most_common_color, most_common_count = color_counts.most_common(1)[0]
+            total_samples = len(colors)
+            
+            # 计算众数颜色占比
+            ratio = most_common_count / total_samples
+            
+            # 如果众数颜色占比小于50%，认为背景杂乱
+            is_messy = ratio < 0.5
+            
+            return most_common_color, is_messy
+        
+        # 默认返回白色，非杂乱
+        return (255, 255, 255), False
+
+    def line_scan_remove_background(self, image: Image.Image, bg_color: tuple, tolerance: int = 30, min_consecutive: int = 5) -> Image.Image:
+        """逐行扫描去除连续的背景色区域"""
+        # 转换为RGBA模式
+        if image.mode != 'RGBA':
+            image = image.convert('RGBA')
+        
+        # 转换为numpy数组
+        data = np.array(image)
+        width, height = image.size
+        
+        def color_distance(c1, c2):
+            """计算颜色距离"""
+            return np.sqrt(sum((a - b) ** 2 for a, b in zip(c1[:3], c2[:3])))
+        
+        def is_background_color(pixel_color):
+            """判断是否为背景色"""
+            return color_distance(pixel_color[:3], bg_color) <= tolerance
+        
+        def process_line(line_data, is_horizontal=True):
+            """处理一行或一列的数据，去除连续的背景色区域"""
+            line_length = len(line_data)
+            i = 0
+            
+            while i < line_length:
+                # 检查当前像素是否为背景色
+                if is_background_color(line_data[i]):
+                    # 找到连续背景色区域的结束位置
+                    consecutive_start = i
+                    while i < line_length and is_background_color(line_data[i]):
+                        i += 1
+                    consecutive_end = i
+                    consecutive_length = consecutive_end - consecutive_start
+                    
+                    # 如果连续背景色区域超过阈值，设置为透明
+                    if consecutive_length > min_consecutive:
+                        for j in range(consecutive_start, consecutive_end):
+                            line_data[j] = (0, 0, 0, 0)  # 设置为透明
+                else:
+                    i += 1
+            
+            return line_data
+        
+        # 逐行扫描（水平方向）
+        print(f"    🔍 逐行扫描（水平方向）...")
+        for y in range(height):
+            row_data = data[y, :].copy()
+            processed_row = process_line(row_data, is_horizontal=True)
+            data[y, :] = processed_row
+        
+        # 逐列扫描（垂直方向）
+        print(f"    🔍 逐列扫描（垂直方向）...")
+        for x in range(width):
+            col_data = data[:, x].copy()
+            processed_col = process_line(col_data, is_horizontal=False)
+            data[:, x] = processed_col
+        
+        # 转换回PIL图像
+        return Image.fromarray(data, 'RGBA')
+
+    def crop_transparent_borders(self, image: Image.Image) -> Image.Image:
+        """裁剪透明边界，去除多余区域"""
+        if image.mode != 'RGBA':
+            return image
+        
+        # 转换为numpy数组
+        data = np.array(image)
+        
+        # 获取alpha通道
+        alpha = data[:, :, 3]
+        
+        # 找到非透明像素的边界
+        non_transparent = np.where(alpha > 0)
+        
+        if len(non_transparent[0]) == 0:
+            # 如果图像完全透明，返回最小尺寸
+            return image.crop((0, 0, 1, 1))
+        
+        # 计算边界框
+        min_y, max_y = non_transparent[0].min(), non_transparent[0].max()
+        min_x, max_x = non_transparent[1].min(), non_transparent[1].max()
+        
+        # 添加小的边距（5像素）
+        padding = 5
+        width, height = image.size
+        
+        min_x = max(0, min_x - padding)
+        min_y = max(0, min_y - padding)
+        max_x = min(width - 1, max_x + padding)
+        max_y = min(height - 1, max_y + padding)
+        
+        # 裁剪图像
+        cropped = image.crop((min_x, min_y, max_x + 1, max_y + 1))
+        
+        return cropped
+
+    def post_process_image(self, image: Image.Image) -> Image.Image:
+        """后处理图像：去除背景并裁剪多余区域，如果背景杂乱则返回None"""
+        print(f"    🔧 后处理图像...")
+        
+        # 检测背景颜色和杂乱程度
+        bg_color, is_messy = self.detect_background_color(image)
+        
+        if is_messy:
+            print(f"    ❌ 检测到杂乱背景，抛弃此图片")
+            return None
+        
+        print(f"    📊 检测到背景色: {bg_color}")
+        
+        # 使用逐行扫描去除背景
+        processed_image = self.line_scan_remove_background(image, bg_color, tolerance=30, min_consecutive=5)
+        
+        # 裁剪透明边界
+        cropped_image = self.crop_transparent_borders(processed_image)
+        
+        original_size = image.size
+        final_size = cropped_image.size
+        print(f"    ✂️ 尺寸调整: {original_size} → {final_size}")
+        
+        return cropped_image
+
+    def generate_prompt_and_image(self, concept: str, topic: str, keyword: str, category: str) -> str:
+        """为单个概念生成prompt并生成图像"""
+        print(f"  🎨 处理: {concept[:50]}...")
+        
+        # 生成设计prompt
+        prompt = self.design_prompt_template.format(
+            topic=topic,
+            style_keyword=keyword,
+            concept=concept
+        )
+        
+        response = self.openai_client.chat.completions.create(
+            model="gpt-5-mini",
+            messages=[
+                {"role": "user", "content": prompt}
+            ],
+            temperature=0.7
+        )
+        
+        image_prompt = response.choices[0].message.content.strip()
+        
+        # 生成图像使用imagen-4.0
+        response = self.genai_client.models.generate_images(
+            model='imagen-4.0-fast-generate-001',
+            prompt=image_prompt,
+            config=types.GenerateImagesConfig(
+                number_of_images=1,
+                aspect_ratio="1:1",
+            )
+        )
+        
+        # 保存图像
+        if response and hasattr(response, 'generated_images') and response.generated_images:
+            generated_image = response.generated_images[0]
+            image = Image.open(BytesIO(generated_image.image.image_bytes))
+            
+            # 后处理图像：去除背景
+            processed_image = self.post_process_image(image)
+            
+            # 如果图像被抛弃（杂乱背景），返回None
+            if processed_image is None:
+                print(f"    🗑️ 图片已抛弃")
+                return None
+            
+            # 构建文件名 - 使用连字符连接，下划线替换空格
+            safe_topic = topic.replace(' ', '_')
+            safe_category = category.replace(' ', '_')
+            safe_concept = concept[:30].replace(' ', '_')
+            
+            # 移除非字母数字和允许的字符
+            safe_topic = "".join(c for c in safe_topic if c.isalnum() or c in ('_', '-')).strip('_-')
+            safe_category = "".join(c for c in safe_category if c.isalnum() or c in ('_', '-')).strip('_-')
+            safe_concept = "".join(c for c in safe_concept if c.isalnum() or c in ('_', '-')).strip('_-')
+            
+            timestamp = int(time.time())
+            filename = f"{safe_topic}-{safe_category}-{safe_concept}-{timestamp}.png"
+            filepath = os.path.join(self.output_dir, filename)
+            
+            processed_image.save(filepath)
+            print(f"    ✅ 保存: {os.path.basename(filepath)}")
+            return filepath
+        
+        return None
+
+    def run_pipeline(self) -> Dict:
+        """运行完整的批量生成pipeline"""
+        print("🚀 开始图像批量生成Pipeline")
+        
+        # 1. 随机选择category、keyword和topic
+        category, keyword, topic = self.select_random_category_and_elements()
+        
+        # 2. 生成10个概念
+        concepts = self.generate_concepts(topic)
+        
+        # 3. 评判并选择5个最佳概念
+        selected_concepts = self.evaluate_and_select_concepts(concepts)
+        
+        # 4. 并行生成prompt和图像
+        print(f"🖼️ 并行生成 {len(selected_concepts)} 张图像...")
+        generated_files = []
+        
+        with ThreadPoolExecutor(max_workers=3) as executor:
+            futures = []
+            for concept in selected_concepts:
+                future = executor.submit(
+                    self.generate_prompt_and_image, 
+                    concept, topic, keyword, category
+                )
+                futures.append(future)
+            
+            for future in futures:
+                result = future.result()
+                if result:  # 只有成功生成且未被抛弃的图片才会被添加
+                    generated_files.append(result)
+        
+        print(f"✅ 完成! 生成 {len(generated_files)} 张图像")
+        
+        return {
+            'category': category,
+            'keyword': keyword,
+            'topic': topic,
+            'generated_images': len(generated_files),
+            'output_files': generated_files
+        }
+
+
+def main():
+    """主函数"""
+    random.seed(int(time.time()))
+    generator = ImageBatchGenerator()
+    for i in range(10):
+        result = generator.run_pipeline()
+    print(f"📊 结果: {result['generated_images']} 张图像已保存")
+
+
+if __name__ == "__main__":
+    main()

icon_generation/backup/image_generator.py

@@ -0,0 +1,64 @@
+import os
+from openai import OpenAI
+from PIL import Image
+from io import BytesIO
+import base64
+import sys
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+from config import api_key, base_url
+
+# OpenAI API configuration
+API_KEY = api_key
+API_PROVIDER = base_url
+
+client = OpenAI(
+    api_key=API_KEY,
+    base_url=API_PROVIDER,
+)
+
+
+from openai import OpenAI
+import base64
+import os
+
+client = OpenAI(
+    api_key=API_KEY,
+    base_url=API_PROVIDER
+)
+
+result = client.images.generate(
+    model="gpt-image-1",
+    prompt=prompt,
+    n=1, # 单次出图数量，最多 10 张
+    size="1024x1024", # 1024x1024 (square), 1536x1024 (3:2 landscape), 1024x1536 (2:3 portrait), auto (default) 
+    quality="low", # high, medium, low, auto (default)
+    moderation="low", # low, auto (default) 需要升级 openai 包 📍
+    background="auto", # transparent, opaque, auto (default)
+)
+
+
+design_prompt_template = """
+[TASK START]
+OBJECTIVE: Generate a text-to-image prompt for a single, isolated clipart icon based on the provided inputs.
+
+INPUTS:
+Topic: TOPIC_PLACEHOLDER
+Style Keyword: STYLE_KEYWORD_PLACEHOLDER
+
+PROCESS:
+
+Conceptualize: First, determine a single, specific, and universally recognizable visual concept (a concrete object or scene) that best represents the broad Topic.
+Generate: Second, write a text-to-image prompt describing this visual concept, rendered using the specified Style Keyword.
+
+CONSTRAINTS:
+
+The output must be a single icon or a small, unified group of objects.
+The icon MUST be isolated on a plain white background.
+The final prompt must be concise and descriptive.
+
+REQUIRED OUTPUT: (Provide your answer in this exact structure)
+Selected Concept: [The specific visual metaphor you chose for the topic]
+Image Prompt: [The final text-to-image prompt]
+
+[TASK END]
+"""

icon_generation/backup/json_to_txt.py

@@ -0,0 +1,77 @@
+#!/usr/bin/env python3
+"""
+将 refined_domains.json 转换为 txt 格式
+每行包含: domain, specific_value, value_count
+按 value_count 降序排序
+"""
+
+import json
+
+
+def convert_json_to_txt(input_file: str, output_file: str):
+    """
+    转换 JSON 到 TXT 格式
+    
+    Args:
+        input_file: 输入的 JSON 文件
+        output_file: 输出的 TXT 文件
+    """
+    print(f"📖 读取文件: {input_file}")
+    
+    # 读取 JSON 数据
+    with open(input_file, 'r', encoding='utf-8') as f:
+        data = json.load(f)
+    
+    print(f"✅ 成功读取 {len(data)} 个 domains")
+    
+    # 收集所有的 domain-value-count 三元组
+    all_pairs = []
+    
+    for item in data:
+        domain = item['domain']
+        for value_info in item['values']:
+            value = value_info['value']
+            count = value_info['count']
+            all_pairs.append((domain, value, count))
+    
+    print(f"📊 总共收集 {len(all_pairs)} 个 domain-value pairs")
+    
+    # 按 count 降序排序
+    print("🔄 按 value_count 降序排序...")
+    all_pairs.sort(key=lambda x: x[2], reverse=True)
+    
+    # 写入 TXT 文件
+    print(f"💾 写入文件: {output_file}")
+    
+    with open(output_file, 'w', encoding='utf-8') as f:
+        for domain, value, count in all_pairs:
+            # 确保 value 中的逗号不会破坏 CSV 格式
+            # 如果 value 包含逗号，用引号包裹
+            if ',' in value:
+                value = f'"{value}"'
+            if ',' in domain:
+                domain = f'"{domain}"'
+            
+            f.write(f"{domain},{value},{count}\n")
+    
+    print(f"✅ 转换完成！")
+    print(f"\n📈 统计信息:")
+    print(f"   总 pairs 数: {len(all_pairs):,}")
+    print(f"   最高 count: {all_pairs[0][2]:,} ({all_pairs[0][0]}: {all_pairs[0][1]})")
+    print(f"   最低 count: {all_pairs[-1][2]:,} ({all_pairs[-1][0]}: {all_pairs[-1][1]})")
+    
+    # 显示前 10 个
+    print(f"\n🏆 Top 10 pairs:")
+    for i, (domain, value, count) in enumerate(all_pairs[:10], 1):
+        print(f"   {i:2d}. {domain}: {value} ({count:,})")
+
+
+if __name__ == '__main__':
+    input_file = 'refined_domains.json'
+    output_file = 'domain_value_pairs.txt'
+    
+    convert_json_to_txt(input_file, output_file)
+
+
+
+

icon_generation/backup/refine_domains.py

@@ -0,0 +1,537 @@
+#!/usr/bin/env python3
+"""
+使用 LLM 对 filtered.json 中的 name 字段进行 refine，
+将其改为更明确的 domain，并按 count 排序
+"""
+
+import json
+import os
+import requests
+from typing import Dict, Optional, List
+from collections import Counter
+from concurrent.futures import ThreadPoolExecutor, as_completed
+import threading
+
+
+class DomainRefiner:
+    """使用 LLM 来 refine domain names"""
+    
+    def __init__(self, api_key=None, base_url=None, model=None):
+        """
+        初始化 LLM analyzer
+        
+        Args:
+            api_key: API key
+            base_url: API base URL
+            model: Model name
+        """
+        self.api_key = api_key or os.getenv("OPENAI_API_KEY") or os.getenv("AIHUBMIX_API_KEY", "")
+        self.base_url = base_url or os.getenv("OPENAI_BASE_URL", "https://aihubmix.com/v1")
+        self.model = model or os.getenv("OPENAI_MODEL", "gemini-2.5-flash")
+        self.lock = threading.Lock()  # 线程锁，用于打印
+    
+    def filter_values(self, domain_name: str, values_with_counts: List[Dict]) -> List[Dict]:
+        """
+        使用 LLM 过滤掉不属于该 domain 的 specific values 和相似/重复的 values
+        
+        Args:
+            domain_name: domain 名称
+            values_with_counts: 包含 value 和 count 的列表 [{"value": "...", "count": ...}, ...]
+        
+        Returns:
+            过滤后的 values 列表
+        """
+        # 如果值太多，只发送 top 50 给 LLM
+        values_to_check = values_with_counts[:50] if len(values_with_counts) > 50 else values_with_counts
+        
+        prompt = self._build_filter_prompt(domain_name, values_to_check)
+        
+        try:
+            response = self._query_llm(prompt)
+            
+            if response:
+                # 清理可能的 markdown 代码块
+                cleaned_response = response.strip()
+                if cleaned_response.startswith('```'):
+                    lines = cleaned_response.split('\n')
+                    cleaned_response = '\n'.join(lines[1:-1] if lines[-1].strip() == '```' else lines[1:])
+                    cleaned_response = cleaned_response.replace('```json', '').replace('```', '').strip()
+                
+                result = json.loads(cleaned_response)
+                
+                # 验证返回格式
+                if 'filtered_values' in result:
+                    filtered_values_set = set(result['filtered_values'])
+                    
+                    # 对于 top 50，过滤它们
+                    filtered_top = [v for v in values_to_check if v['value'] in filtered_values_set]
+                    
+                    # 如果原始列表更长，保留剩余的（因为没有检查）
+                    if len(values_with_counts) > 50:
+                        remaining = values_with_counts[50:]
+                        filtered_list = filtered_top + remaining
+                    else:
+                        filtered_list = filtered_top
+                    
+                    return filtered_list
+                else:
+                    with self.lock:
+                        print(f"      ⚠️  LLM 过滤响应缺少字段，保留原始值")
+                    return values_with_counts
+                    
+            else:
+                with self.lock:
+                    print(f"      ⚠️  LLM 过滤失败，保留原始值")
+                return values_with_counts
+                
+        except json.JSONDecodeError as e:
+            with self.lock:
+                print(f"      ⚠️  LLM 过滤响应不是有效的 JSON，保留原始值")
+            return values_with_counts
+        except Exception as e:
+            with self.lock:
+                print(f"      ⚠️  值过滤错误: {e}，保留原始值")
+            return values_with_counts
+    
+    def _build_filter_prompt(self, domain_name: str, values_with_counts: List[Dict]) -> str:
+        """构建值过滤的 prompt"""
+        
+        values_str = '\n'.join([f'  - "{v["value"]}" (count: {v["count"]})' for v in values_with_counts])
+        
+        prompt = f"""Given a domain name and its associated values, please filter out:
+1. Values that don't truly belong to this domain (too specific, off-topic, or irrelevant)
+2. Similar or duplicate values (keep the most common or representative one)
+3. Values that are too generic or ambiguous
+
+Domain: "{domain_name}"
+
+Values to filter:
+{values_str}
+
+Please analyze these values and return ONLY the values that:
+- Clearly belong to this domain
+- Are distinct (not duplicates or very similar)
+- Are meaningful attributes
+
+Return your response in the following JSON format ONLY (no markdown, no extra text):
+{{
+    "filtered_values": ["value1", "value2", ...],
+    "removed_count": number_of_removed_values,
+    "reasoning": "Brief explanation of filtering criteria used"
+}}
+
+Examples:
+- Domain "Movie Genre" with values ["Action", "action movie", "ACT"] → Keep only "Action"
+- Domain "Country" with values ["USA", "New York", "California"] → Remove "New York", "California" (cities, not countries)
+"""
+        
+        return prompt
+    
+    def refine_domain_name(self, name: str, sample_values: List[str], total_count: int) -> Dict:
+        """
+        使用 LLM 分析并 refine domain name
+        
+        Args:
+            name: 原始 name
+            sample_values: 一些示例 values
+            total_count: 总计数
+        
+        Returns:
+            {
+                'original_name': str,
+                'refined_domain': str,
+                'reasoning': str
+            }
+        """
+        prompt = self._build_domain_refinement_prompt(name, sample_values, total_count)
+        
+        try:
+            response = self._query_llm(prompt)
+            
+            if response:
+                # 清理可能的 markdown 代码块
+                cleaned_response = response.strip()
+                if cleaned_response.startswith('```'):
+                    lines = cleaned_response.split('\n')
+                    cleaned_response = '\n'.join(lines[1:-1] if lines[-1].strip() == '```' else lines[1:])
+                    cleaned_response = cleaned_response.replace('```json', '').replace('```', '').strip()
+                
+                result = json.loads(cleaned_response)
+                
+                # 验证返回格式
+                if 'refined_domain' in result:
+                    result['original_name'] = name
+                    return result
+                else:
+                    with self.lock:
+                        print(f"      ❌ LLM 响应缺少必需字段: {result}")
+                    return None
+                    
+            else:
+                with self.lock:
+                    print("      ❌ LLM API 调用失败")
+                return None
+                
+        except json.JSONDecodeError as e:
+            with self.lock:
+                print(f"      ❌ LLM 响应不是有效的 JSON: {e}")
+                print(f"      响应内容: {response[:500]}...")
+            return None
+        except Exception as e:
+            with self.lock:
+                print(f"      ❌ Domain refinement 错误: {e}")
+            return None
+    
+    def _build_domain_refinement_prompt(self, name: str, sample_values: List[str], total_count: int) -> str:
+        """构建 domain refinement 的 prompt"""
+        
+        sample_values_str = ', '.join(f'"{v}"' for v in sample_values[:10])
+        
+        prompt = f"""Given a data field name and its sample values, please refine the name to a more specific and clear domain name.
+
+The domain name should:
+1. Clearly indicate what category/dimension this field represents
+2. Be consistent and professional
+3. Form a clear "domain: specific attribute" relationship with its values
+4. Be concise (1-3 words)
+
+Input Information:
+- Original Field Name: "{name}"
+- Sample Values: {sample_values_str}
+- Total Entries Count: {total_count}
+
+Please analyze the field name and sample values, then provide:
+1. A refined domain name that better describes this dimension
+2. Brief reasoning for your choice
+
+Return your response in the following JSON format ONLY (no markdown, no extra text):
+{{
+    "refined_domain": "YourRefinedDomainName",
+    "reasoning": "Brief explanation of why this domain name is more appropriate"
+}}
+
+Examples:
+- Original: "Genre" with values ["Action", "Rock", "Pop"] → Refined: "Entertainment Genre"
+- Original: "Type" with values ["Movie", "TV Show"] → Refined: "Media Type"
+- Original: "Category" with values ["Electronics", "Books"] → Refined: "Product Category"
+"""
+        
+        return prompt
+    
+    def _query_llm(self, prompt: str) -> Optional[str]:
+        """
+        查询 LLM API
+        
+        Args:
+            prompt: 发送给 LLM 的 prompt
+        
+        Returns:
+            str: LLM 响应内容
+        """
+        headers = {
+            'Authorization': f'Bearer {self.api_key}',
+            'Content-Type': 'application/json'
+        }
+        
+        data = {
+            'model': self.model,
+            'messages': [
+                {
+                    'role': 'system', 
+                    'content': 'You are a data modeling expert specialized in creating clear, semantic domain names. Always return valid JSON format only, without any markdown formatting or extra text.'
+                },
+                {
+                    'role': 'user', 
+                    'content': prompt
+                }
+            ],
+            'temperature': 0.3
+        }
+        
+        try:
+            response = requests.post(
+                f'{self.base_url}/chat/completions',
+                headers=headers,
+                json=data,
+                timeout=30
+            )
+            response.raise_for_status()
+            
+            result = response.json()
+            return result['choices'][0]['message']['content'].strip()
+            
+        except requests.exceptions.Timeout:
+            with self.lock:
+                print("❌ LLM API 超时")
+            return None
+        except requests.exceptions.HTTPError as e:
+            with self.lock:
+                print(f"❌ LLM API HTTP 错误: {e}")
+                if hasattr(e.response, 'text'):
+                    print(f"   响应: {e.response.text[:200]}")
+            return None
+        except requests.exceptions.RequestException as e:
+            with self.lock:
+                print(f"❌ LLM API 请求错误: {e}")
+            return None
+        except KeyError as e:
+            with self.lock:
+                print(f"❌ LLM API 响应格式错误: {e}")
+            return None
+
+
+def process_single_item(item: Dict, idx: int, total: int, refiner: DomainRefiner, start_time: float, start_idx: int) -> Dict:
+    """
+    处理单个数据项
+    
+    Args:
+        item: 数据项
+        idx: 当前索引
+        total: 总数量
+        refiner: DomainRefiner 实例
+        start_time: 开始时间
+        start_idx: 起始索引
+    
+    Returns:
+        处理后的数据项
+    """
+    import time
+    
+    # 计算进度信息
+    progress = (idx + 1) / total * 100
+    elapsed = time.time() - start_time
+    avg_time = elapsed / (idx - start_idx + 1) if idx > start_idx else 0
+    remaining = avg_time * (total - idx - 1)
+    
+    with refiner.lock:
+        print(f"\n{'=' * 80}")
+        print(f"🔄 处理进度: {idx + 1}/{total} ({progress:.1f}%)")
+        print(f"⏱️  已用时间: {elapsed:.1f}秒 | 预计剩余: {remaining:.1f}秒")
+        print(f"📝 当前字段: {item['name']}")
+        print(f"   - 值数量: {item['num_values']}")
+        print(f"   - 总计数: {item['total_count']}")
+    
+    # 提取 sample values
+    sample_values = [v['value'] for v in item['values'][:15]]
+    
+    with refiner.lock:
+        print(f"   - 示例值: {', '.join(sample_values[:5])}")
+        print(f"🤖 调用 LLM 进行 domain refinement...")
+    
+    # 使用 LLM refine domain name
+    refined_result = refiner.refine_domain_name(
+        item['name'],
+        sample_values,
+        item['total_count']
+    )
+    
+    if refined_result:
+        refined_domain = refined_result['refined_domain']
+        reasoning = refined_result.get('reasoning', '')
+        
+        with refiner.lock:
+            print(f"✅ Domain refinement 成功!")
+            print(f"   原始名称: '{item['name']}'")
+            print(f"   优化域名: '{refined_domain}'")
+            print(f"   优化理由: {reasoning}")
+            print(f"🔍 调用 LLM 进行 value filtering...")
+        
+        # 过滤 values
+        filtered_values = refiner.filter_values(refined_domain, item['values'])
+        
+        with refiner.lock:
+            removed_count = len(item['values']) - len(filtered_values)
+            print(f"✅ Value filtering 完成!")
+            print(f"   原始值数量: {len(item['values'])}")
+            print(f"   过滤后数量: {len(filtered_values)}")
+            print(f"   移除数量: {removed_count}")
+        
+        # 构建新的数据项
+        refined_item = {
+            'domain': refined_domain,
+            'original_name': item['name'],
+            'num_values': len(filtered_values),
+            'original_num_values': item['num_values'],
+            'total_count': sum(v['count'] for v in filtered_values),
+            'original_total_count': item['total_count'],
+            'values': filtered_values,
+            'refinement_reasoning': reasoning
+        }
+        
+        return refined_item
+    else:
+        # 如果 LLM 失败，保留原始 name 作为 domain
+        with refiner.lock:
+            print(f"⚠️  LLM refinement 失败，使用原始 name，跳过值过滤")
+        
+        refined_item = {
+            'domain': item['name'],
+            'original_name': item['name'],
+            'num_values': item['num_values'],
+            'original_num_values': item['num_values'],
+            'total_count': item['total_count'],
+            'original_total_count': item['total_count'],
+            'values': item['values'],
+            'refinement_reasoning': 'LLM refinement failed, kept original'
+        }
+        
+        return refined_item
+
+
+def process_filtered_json(input_file: str, output_file: str, temp_file: str = None, num_threads: int = 10):
+    """
+    处理 filtered.json 文件（并行处理）
+    
+    Args:
+        input_file: 输入文件路径
+        output_file: 输出文件路径
+        temp_file: 临时文件路径，用于实时保存中间结果
+        num_threads: 并行线程数
+    """
+    import os
+    import time
+    
+    if temp_file is None:
+        temp_file = output_file.replace('.json', '_temp.json')
+    
+    print(f"📖 读取文件: {input_file}")
+    print(f"💾 临时文件: {temp_file}")
+    print(f"✨ 最终文件: {output_file}")
+    print(f"🔧 并行线程数: {num_threads}")
+    
+    # 读取原始数据
+    with open(input_file, 'r', encoding='utf-8') as f:
+        data = json.load(f)
+    
+    print(f"✅ 成功读取 {len(data)} 个字段\n")
+    print("=" * 80)
+    
+    # 检查是否已有临时文件（支持断点续传）
+    refined_data = []
+    start_idx = 0
+    processed_names = set()
+    
+    if os.path.exists(temp_file):
+        print(f"📂 发现临时文件，尝试恢复进度...")
+        try:
+            with open(temp_file, 'r', encoding='utf-8') as f:
+                refined_data = json.load(f)
+            start_idx = len(refined_data)
+            processed_names = {item['original_name'] for item in refined_data}
+            print(f"✅ 已恢复 {start_idx} 个字段的处理结果")
+        except Exception as e:
+            print(f"⚠️  临时文件读取失败: {e}，从头开始")
+            refined_data = []
+            start_idx = 0
+            processed_names = set()
+    
+    # 过滤掉已处理的项
+    items_to_process = [item for item in data if item['name'] not in processed_names]
+    
+    if not items_to_process:
+        print("✅ 所有项目已处理完成！")
+        return
+    
+    print(f"📋 待处理项目: {len(items_to_process)} 个")
+    print("=" * 80)
+    
+    # 初始化 LLM refiner
+    refiner = DomainRefiner()
+    
+    # 记录开始时间
+    start_time = time.time()
+    
+    # 使用线程池并行处理
+    with ThreadPoolExecutor(max_workers=num_threads) as executor:
+        # 提交所有任务
+        future_to_idx = {
+            executor.submit(
+                process_single_item, 
+                item, 
+                start_idx + i, 
+                len(data), 
+                refiner, 
+                start_time, 
+                start_idx
+            ): (start_idx + i, item)
+            for i, item in enumerate(items_to_process)
+        }
+        
+        # 按完成顺序收集结果
+        completed = 0
+        for future in as_completed(future_to_idx):
+            idx, item = future_to_idx[future]
+            try:
+                result = future.result()
+                refined_data.append(result)
+                completed += 1
+                
+                # 每处理 5 个项目保存一次
+                if completed % 5 == 0:
+                    with refiner.lock:
+                        print(f"\n{'=' * 80}")
+                        print(f"💾 保存中间结果... (已完成 {completed}/{len(items_to_process)})")
+                    with open(temp_file, 'w', encoding='utf-8') as f:
+                        json.dump(refined_data, f, ensure_ascii=False, indent=2)
+                    with refiner.lock:
+                        print(f"✅ 已保存 {len(refined_data)} 个字段到临时文件")
+                
+            except Exception as e:
+                with refiner.lock:
+                    print(f"\n❌ 处理项目 {idx} ({item['name']}) 时出错: {e}")
+    
+    # 最终保存一次
+    print(f"\n{'=' * 80}")
+    print(f"💾 保存最终临时结果...")
+    with open(temp_file, 'w', encoding='utf-8') as f:
+        json.dump(refined_data, f, ensure_ascii=False, indent=2)
+    
+    # 按 total_count 降序排序
+    print(f"\n{'=' * 80}")
+    print("📊 按 total_count 进行排序...")
+    refined_data.sort(key=lambda x: x['total_count'], reverse=True)
+    
+    # 保存最终结果
+    print(f"💾 保存最终结果到: {output_file}")
+    with open(output_file, 'w', encoding='utf-8') as f:
+        json.dump(refined_data, f, ensure_ascii=False, indent=2)
+    
+    print(f"\n{'=' * 80}")
+    print(f"✅ 处理完成！共处理 {len(refined_data)} 个字段")
+    print(f"⏱️  总耗时: {time.time() - start_time:.1f}秒")
+    
+    # 输出统计信息
+    print(f"\n{'=' * 80}")
+    print("📈 统计信息:")
+    print(f"   总字段数: {len(refined_data)}")
+    print(f"   总记录数（过滤后）: {sum(item['total_count'] for item in refined_data):,}")
+    print(f"   总记录数（原始）: {sum(item.get('original_total_count', item['total_count']) for item in refined_data):,}")
+    
+    # 计算过滤统计
+    total_values_before = sum(item.get('original_num_values', item['num_values']) for item in refined_data)
+    total_values_after = sum(item['num_values'] for item in refined_data)
+    print(f"   总值数量（原始）: {total_values_before:,}")
+    print(f"   总值数量（过滤后）: {total_values_after:,}")
+    print(f"   过滤比例: {(1 - total_values_after/total_values_before)*100:.1f}%")
+    
+    # 显示前 10 个 domain
+    print(f"\n{'=' * 80}")
+    print("🏆 Top 10 Domains (按 total_count):")
+    for i, item in enumerate(refined_data[:10]):
+        print(f"\n   {i+1}. {item['domain']} (原: {item['original_name']})")
+        print(f"      Count: {item['total_count']:,}, Values: {item['num_values']}")
+        print(f"      示例值: {', '.join([v['value'] for v in item['values'][:5]])}")
+    
+    # 删除临时文件
+    if os.path.exists(temp_file):
+        print(f"\n🗑��  保留临时文件以备恢复: {temp_file}")
+        # os.remove(temp_file)  # 暂时不删除，以便需要时恢复
+
+
+if __name__ == '__main__':
+    input_file = '/home/lizhen/ChartPipeline/icon_generation/filtered.json'
+    output_file = '/home/lizhen/ChartPipeline/icon_generation/refined_domains.json'
+    
+    process_filtered_json(input_file, output_file, num_threads=10)
+
+

icon_generation/backup/summarize_domains.py

@@ -0,0 +1,142 @@
+#!/usr/bin/env python3
+"""
+总结 domain_value_pairs_enhanced.txt 中的所有 domains 及其 top 5 attributes
+"""
+
+from collections import defaultdict
+from typing import List, Tuple
+
+
+def parse_csv_line(line: str) -> Tuple[str, str, int]:
+    """解析 CSV 行"""
+    parts = []
+    current = []
+    in_quotes = False
+    
+    for char in line:
+        if char == '"':
+            in_quotes = not in_quotes
+        elif char == ',' and not in_quotes:
+            parts.append(''.join(current).strip())
+            current = []
+        else:
+            current.append(char)
+    parts.append(''.join(current).strip())
+    
+    if len(parts) >= 3:
+        domain = parts[0]
+        attribute = parts[1]
+        freq = int(parts[2])
+        return domain, attribute, freq
+    else:
+        return None, None, 0
+
+
+def summarize_domains(input_file: str, output_file: str = None):
+    """
+    总结所有 domains 及其 top 5 attributes
+    
+    Args:
+        input_file: 输入文件路径
+        output_file: 输出文件路径（可选）
+    """
+    print("=" * 80)
+    print("Domain 和 Top Attributes 汇总")
+    print("=" * 80)
+    print()
+    
+    print(f"📖 读取文件: {input_file}")
+    
+    # 收集每个 domain 的所有 attributes
+    domain_attributes = defaultdict(list)
+    
+    with open(input_file, 'r', encoding='utf-8') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                continue
+            
+            domain, attribute, freq = parse_csv_line(line)
+            if domain:
+                domain_attributes[domain].append((attribute, freq))
+    
+    print(f"✅ 成功读取 {len(domain_attributes)} 个 domains")
+    print()
+    
+    # 对每个 domain 的 attributes 按 frequency 排序
+    for domain in domain_attributes:
+        domain_attributes[domain].sort(key=lambda x: x[1], reverse=True)
+    
+    # 按 domain 的 top attribute 的 frequency 排序 domains
+    sorted_domains = sorted(
+        domain_attributes.items(),
+        key=lambda x: x[1][0][1] if x[1] else 0,
+        reverse=True
+    )
+    
+    # 生成输出
+    print("=" * 80)
+    print("📊 所有 Domains 及 Top 5 Attributes:")
+    print("=" * 80)
+    print()
+    
+    summary_lines = []
+    
+    for idx, (domain, attributes) in enumerate(sorted_domains, 1):
+        # 获取 top 5 attributes
+        top_5 = attributes[:5]
+        top_5_names = [attr for attr, _ in top_5]
+        
+        # 格式化输出
+        line1 = f"{idx}. {domain}"
+        line2 = f"Attributes: {', '.join(top_5_names)}"
+        
+        summary_lines.append(line1)
+        summary_lines.append(line2)
+        summary_lines.append("")  # 空行
+        
+        print(line1)
+        print(line2)
+        print()
+    
+    # 保存到文件（如果指定）
+    if output_file:
+        print("=" * 80)
+        print(f"💾 保存到文件: {output_file}")
+        
+        with open(output_file, 'w', encoding='utf-8') as f:
+            f.write("Domain Summary with Top 5 Attributes\n")
+            f.write("=" * 80 + "\n\n")
+            for line in summary_lines:
+                f.write(line + "\n")
+        
+        print(f"✅ 已保存")
+    
+    # 统计信息
+    print()
+    print("=" * 80)
+    print("📈 统计信息:")
+    print("=" * 80)
+    print(f"总 Domains 数: {len(domain_attributes)}")
+    print(f"总 Attributes 数: {sum(len(attrs) for attrs in domain_attributes.values())}")
+    
+    # 计算每个 domain 的平均 attributes 数
+    avg_attrs = sum(len(attrs) for attrs in domain_attributes.values()) / len(domain_attributes)
+    print(f"平均每个 Domain 的 Attributes 数: {avg_attrs:.1f}")
+    
+    # Domains 按 attributes 数量分布
+    attr_counts = [len(attrs) for attrs in domain_attributes.values()]
+    print(f"Attributes 数量范围: {min(attr_counts)} - {max(attr_counts)}")
+    
+    print()
+    print("=" * 80)
+    print("✨ 汇总完成！")
+    print("=" * 80)
+
+
+if __name__ == '__main__':
+    input_file = 'domain_value_pairs_enhanced.txt'
+    output_file = 'domain_summary.txt'
+    
+    summarize_domains(input_file, output_file)
+

icon_generation/backup/topic_style.json

@@ -0,0 +1,520 @@
+{
+    "flat_and_minimal": {
+      "description": "扁平与极简：最清晰、最现代、最通用的信息图风格，专注于形状和颜色的直接传达。",
+      "keywords": [
+        "flat_solid_icon",
+        "minimal_shape",
+        "solid_color_fill",
+        "negative_space_icon",
+        "modern_minimal_form",
+        "basic_geometric_shape",
+        "essential_silhouette",
+        "flat_ui_style",
+        "simplified_form",
+        "color_block_icon",
+        "app_style_icon",
+        "wayfinding_style",
+        "flat_badge",
+        "corporate_flat",
+        "sharp_edge_vector"
+      ],
+      "topics": [
+        "business information",
+        "economy",
+        "products and services",
+        "online and remote learning",
+        "school",
+        "sustainability",
+        "public health",
+        "government policy",
+        "technology and engineering",
+        "demographics",
+        "social condition",
+        "welfare"
+      ]
+    },
+    "outline_and_line_art": {
+      "description": "描边与线条：使用线条而非填充，营造出更轻盈、更技术感或更优雅的视觉感受。",
+      "keywords": [
+        "monoline_icon",
+        "outline_glyph",
+        "thin_stroke_style",
+        "bold_outline_shape",
+        "line_art_pictogram",
+        "continuous_line_icon",
+        "minimal_stroke_design",
+        "technical_drawing_style",
+        "wireframe_style_icon",
+        "stroked_vector_shape",
+        "rounded_outline_icon",
+        "sharp_edge_line_art",
+        "duoline_style_icon",
+        "minimal_line_icon",
+        "stroke_only_glyph"
+      ],
+      "topics": [
+        "technology and engineering",
+        "scientific research",
+        "biomedical science",
+        "mathematics",
+        "business information",
+        "products and services",
+        "health facility",
+        "medical profession",
+        "scientific institution",
+        "government policy",
+        "architecture"
+      ]
+    },
+    "corporate_and_professional": {
+      "description": "商务与专业：传达信任、精确和可靠性，适用于商业报告、金融数据和技术文档。",
+      "keywords": [
+        "corporate_clean_icon",
+        "professional_minimal",
+        "business_infographic_style",
+        "data_viz_icon_style",
+        "tech_corporate_look",
+        "corporate_soft_gradient",
+        "formal_vector_shape",
+        "corporate_duotone",
+        "abstract_corporate_shape",
+        "clean_tech_icon",
+        "startup_style_icon",
+        "corporate_badge",
+        "enterprise_icon_style",
+        "presentation_icon_style"
+      ],
+      "topics": [
+        "business enterprise",
+        "business information",
+        "economy",
+        "market and exchange",
+        "products and services",
+        "employment",
+        "employment legislation",
+        "labour market",
+        "labour relations",
+        "retirement",
+        "government",
+        "government policy",
+        "law",
+        "health insurance",
+        "private health care",
+        "sport industry",
+        "sports management and ownership"
+      ]
+    },
+    "hand_drawn_and_organic": {
+      "description": "手绘与有机：具有人情味、创意和不完美感，适用于非正式、创意或个性化主题。",
+      "keywords": [
+        "hand_drawn_icon",
+        "sketchy_style_pictogram",
+        "doodle_icon_style",
+        "organic_shape_fill",
+        "imperfect_lines",
+        "chalkboard_style_icon",
+        "crayon_texture_fill",
+        "pencil_sketch_icon",
+        "hand_drawn_outline",
+        "crafty_paper_cut_style",
+        "whimsical_hand_drawn",
+        "marker_stroke_style",
+        "ink_sketch_glyph",
+        "rustic_hatch_fill",
+        "storybook_icon_style",
+        "organic_blob_shape",
+        "playful_hand_drawn"
+      ],
+      "topics": [
+        "arts and entertainment",
+        "culture",
+        "leisure",
+        "lifestyle",
+        "wellness",
+        "social learning",
+        "students",
+        "family",
+        "communities",
+        "social problem",
+        "non-governmental organisation (NGO)",
+        "conservation",
+        "parents group",
+        "values"
+      ]
+    },
+    "playful_and_cute": {
+      "description": "趣味与可爱：有趣、活泼、引人入胜，非常适合教育、社交媒体或轻松的主题。",
+      "keywords": [
+        "playful_rounded_shape",
+        "cute_cartoon_icon",
+        "kawaii_style_glyph",
+        "kid_friendly_vector",
+        "bouncy_form",
+        "fun_mascot_style",
+        "playful_doodle",
+        "toy_like_icon",
+        "playful_corporate_lite",
+        "educational_playful",
+        "soft_bubble_shape",
+        "childlike_drawing_style",
+        "sticker_style_icon",
+        "bubbly_pictogram",
+        "friendly_cartoon_icon",
+        "playful_badge",
+        "cute_minimal",
+        "rounded_corner_style"
+      ],
+      "topics": [
+        "social learning",
+        "school",
+        "students",
+        "parents group",
+        "family",
+        "leisure",
+        "lifestyle",
+        "birthday",
+        "anniversary",
+        "celebrity",
+        "arts and entertainment",
+        "products and services",
+        "religious festival and holiday",
+        "wellness"
+      ]
+    },
+    "tech_and_futuristic": {
+      "description": "科技与未来：现代、前卫、动感，暗示技术、数据、创新和科幻主题。",
+      "keywords": [
+        "tech_glow_outline",
+        "neon_line_art_icon",
+        "futuristic_hud_style",
+        "sci_fi_ui_glyph",
+        "data_stream_lines",
+        "circuit_board_pattern_fill",
+        "tech_minimal_glyph",
+        "digital_glitch_effect_icon",
+        "glossy_tech_icon",
+        "cyberpunk_style_pictogram",
+        "glowing_edge_effect",
+        "holographic_fill_style",
+        "tech_badge",
+        "dark_mode_ui_icon",
+        "gradient_line_art",
+        "plexus_style_lines",
+        "vector_circuit_icon",
+        "modern_tech_glyph",
+        "data_flow_abstract_shape",
+        "digital_network_icon"
+      ],
+      "topics": [
+        "technology and engineering",
+        "cyber warfare",
+        "scientific research",
+        "biomedical science",
+        "natural science",
+        "online and remote learning",
+        "business information",
+        "market and exchange",
+        "products and services",
+        "mass media",
+        "scientific institution"
+      ]
+    },
+    "retro_and_vintage": {
+      "description": "复古与怀旧：唤起过去的时代感，适用于历史数据、时间线或营造特定的怀旧氛围。",
+      "keywords": [
+        "retro_pixel_art_icon",
+        "8_bit_glyph_style",
+        "vintage_badge_design",
+        "retro_cartoon_style",
+        "mid_century_modern_icon",
+        "70s_groovy_style_vector",
+        "50s_atomic_era_shape",
+        "vintage_stamp_effect",
+        "retro_line_art_icon",
+        "woodcut_style_icon",
+        "engraving_hatch_fill",
+        "retro_tech_look",
+        "vintage_label_style",
+        "retro_halftone_fill",
+        "distressed_texture_overlay",
+        "retro_signage_style",
+        "vintage_script_accent"
+      ],
+      "topics": [
+        "culture",
+        "arts and entertainment",
+        "mass media",
+        "post-war reconstruction",
+        "social condition",
+        "leisure",
+        "lifestyle",
+        "products and services",
+        "history"
+      ]
+    },
+    "isometric_and_3d": {
+      "description": "等距与3D：增加深度和空间感，非常适合表现流程、地图、建筑或堆叠概念。",
+      "keywords": [
+        "isometric_vector_block",
+        "3d_icon_style",
+        "orthographic_view_icon",
+        "2.5d_style_pictogram",
+        "isometric_process_flow",
+        "soft_3d_clay_style",
+        "claymorphism_icon",
+        "glossy_3d_web_icon",
+        "vector_voxel_art",
+        "isometric_map_element",
+        "stacked_layers_style",
+        "3d_minimal_shape",
+        "neo_brutalism_3d_icon",
+        "low_poly_vector_icon",
+        "isometric_grid_style",
+        "3d_chart_icon",
+        "soft_ui_3d_style (neumorphism)",
+        "3d_cartoon_vector",
+        "vector_3d_render_style",
+        "soft_shadow_3d"
+      ],
+      "topics": [
+        "business enterprise",
+        "business information",
+        "technology and engineering",
+        "scientific research",
+        "emergency response",
+        "products and services",
+        "online and remote learning",
+        "market and exchange",
+        "logistics",
+        "supply chain",
+        "architecture",
+        "urban planning"
+      ]
+    },
+    "geometric_and_abstract": {
+      "description": "几何与抽象：概念性、模块化，使用基础形状（圆形、方形、三角形）构建图标。",
+      "keywords": [
+        "geometric_shape_build",
+        "abstract_pictogram",
+        "bauhaus_style_icon",
+        "modular_icon_design",
+        "low_poly_flat_icon",
+        "geometric_pattern_fill",
+        "sacred_geometry_lines",
+        "crystal_polygon_shape",
+        "memphis_style_elements",
+        "color_block_composition",
+        "tangram_style_icon",
+        "minimal_geometric_glyph",
+        "abstract_data_icon",
+        "geometric_logo_style",
+        "pattern_based_icon",
+        "triangular_mesh_icon",
+        "circles_and_squares_build",
+        "minimal_abstract_form",
+        "modernist_style_icon",
+        "geometric_badge_design"
+      ],
+      "topics": [
+        "mathematics",
+        "natural science",
+        "biomedical science",
+        "scientific research",
+        "scientific standards",
+        "technology and engineering",
+        "social sciences",
+        "demographics",
+        "economy",
+        "market and exchange",
+        "data visualization"
+      ]
+    },
+    "textured_and_grainy": {
+      "description": "纹理与颗粒：为扁平图标添加触感和深度，使用颗粒、纸张或喷漆等纹理。",
+      "keywords": [
+        "grain_texture_fill",
+        "noise_overlay_effect",
+        "spray_paint_texture",
+        "stipple_effect_fill",
+        "brushed_texture_icon",
+        "paper_texture_background",
+        "textured_shadow",
+        "craft_paper_style",
+        "flat_with_grain_shading",
+        "distressed_vector_icon",
+        "risograph_texture_effect",
+        "grunge_icon_style",
+        "canvas_texture_overlay",
+        "minimal_texture_accent",
+        "subtle_grain_overlay",
+        "sand_texture_fill",
+        "screen_print_effect_icon",
+        "sponge_paint_texture",
+        "chalky_texture"
+      ],
+      "topics": [
+        "arts and entertainment",
+        "culture",
+        "conservation",
+        "nature",
+        "lifestyle",
+        "leisure",
+        "products and services",
+        "food and drink",
+        "crafts",
+        "social condition"
+      ]
+    },
+    "eco_and_nature": {
+      "description": "生态与自然：强调有机、可持续和自然主题，常使用大地色系和流畅线条。",
+      "keywords": [
+        "organic_line_art_icon",
+        "botanical_icon_style",
+        "leaf_motif_icon",
+        "natural_form_shape",
+        "eco_friendly_badge",
+        "recycled_paper_texture_fill",
+        "hand_drawn_nature_icon",
+        "organic_flowing_lines",
+        "environmental_glyph",
+        "sustainable_icon",
+        "floral_line_art",
+        "woodgrain_pattern_fill",
+        "water_ripple_effect",
+        "soft_natural_shape",
+        "plant_silhouette_icon",
+        "outdoor_adventure_style",
+        "rustic_eco_icon"
+      ],
+      "topics": [
+        "climate change",
+        "conservation",
+        "environmental pollution",
+        "natural resource",
+        "nature",
+        "sustainability",
+        "wellness",
+        "lifestyle",
+        "government policy",
+        "leisure",
+        "welfare",
+        "agriculture"
+      ]
+    },
+    "health_and_wellness": {
+      "description": "健康与保健：干净、平静、柔和的风格，用于医疗、正念和生活方式等主题。",
+      "keywords": [
+        "clean_medical_icon",
+        "soft_rounded_shapes",
+        "calm_minimal_style",
+        "medical_glyph_style",
+        "heartbeat_line_art_icon",
+        "organic_wellness_shape",
+        "minimal_health_glyph",
+        "line_art_anatomy_icon",
+        "soft_gradient_fill_icon",
+        "pill_shape_design",
+        "yoga_pose_silhouette",
+        "mindfulness_icon_style",
+        "zen_style_brush_stroke",
+        "clinical_clean_line",
+        "medical_caduceus_style",
+        "spa_and_relaxation_icon",
+        "healthy_food_pictogram",
+        "scientific_clean_icon"
+      ],
+      "topics": [
+        "disease and condition",
+        "government health care",
+        "health facility",
+        "health insurance",
+        "health organisation",
+        "health treatment and procedure",
+        "medical profession",
+        "private health care",
+        "public health",
+        "wellness",
+        "lifestyle",
+        "biomedical science",
+        "social problem",
+        "family",
+        "leisure"
+      ]
+    },
+    "festive_and_celebratory": {
+      "description": "节日与庆典：明亮、欢快、有趣，用于假日、派对、公告和活动主题。",
+      "keywords": [
+        "confetti_pattern_fill",
+        "celebration_icon",
+        "party_doodle_style",
+        "holiday_glyph_set",
+        "sparkle_and_shine_accent",
+        "ribbon_and_banner_style",
+        "decorative_flourish_icon",
+        "bright_gradient_fill",
+        "carnival_style_pictogram",
+        "festive_badge_design",
+        "birthday_icon_style",
+        "event_pictogram_style",
+        "fireworks_burst_icon",
+        "playful_holiday_cartoon",
+        "invitation_style_glyph",
+        "gold_foil_accent",
+        "retro_party_style",
+        "seasonal_icon_pack",
+        "celebratory_burst_shape"
+      ],
+      "topics": [
+        "anniversary",
+        "award and prize",
+        "birthday",
+        "celebrity",
+        "ceremony",
+        "religious festival and holiday",
+        "sport event",
+        "sport achievement",
+        "leisure",
+        "lifestyle",
+        "family",
+        "communities",
+        "culture",
+        "record and achievement"
+      ]
+    },
+    "pop_art_and_comic": {
+      "description": "波普与漫画：大胆、醒目、高对比度，使用半色调圆点和粗黑轮廓等漫画技巧。",
+      "keywords": [
+        "pop_art_style_icon",
+        "comic_book_pictogram",
+        "halftone_dot_fill",
+        "bold_black_outline",
+        "dynamic_action_lines",
+        "ben_day_dots_fill",
+        "comic_speech_bubble_icon",
+        "pop_art_explosion_shape",
+        "graphic_high_contrast",
+        "retro_comic_style_icon",
+        "warhol_inspired_icon",
+        "screen_print_look_icon",
+        "comic_hatching_lines_fill",
+        "sticker_style_pop_art",
+        "bold_graphic_shape",
+        "pop_art_shadow_style",
+        "graphic_poster_style",
+        "cartoon_pop_art"
+      ],
+      "topics": [
+        "arts and entertainment",
+        "culture",
+        "mass media",
+        "celebrity",
+        "leisure",
+        "lifestyle",
+        "social problem",
+        "products and services",
+        "advertising",
+        "civil unrest",
+        "social condition"
+      ]
+    }
+  }

icon_generation/batch_icon_generator.py

@@ -0,0 +1,267 @@
+import os
+import json
+from google import genai
+from google.genai import types
+from pathlib import Path
+import time
+from multiprocessing import Pool
+from functools import partial
+
+# API配置
+API_KEY = os.getenv("GEMINI_API_KEY") or os.getenv("OPENAI_API_KEY", "")
+client = genai.Client(
+    api_key=API_KEY,
+    http_options={"base_url": os.getenv("GEMINI_BASE_URL", "https://aihubmix.com/gemini")},
+)
+
+# 配置参数
+ASPECT_RATIO = "3:2"
+BATCH_SIZE = 24
+MIN_BATCH_SIZE = 6
+OUTPUT_DIR = "generated_icons"
+NUM_PROCESSES = 3
+
+def load_domain_attributes(file_path):
+    """读取domain_attributes.txt文件，返回扁平化的(domain, attribute)对列表"""
+    domain_attribute_pairs = []
+    current_domain = None
+    
+    with open(file_path, 'r', encoding='utf-8') as f:
+        lines = f.readlines()
+    
+    for line in lines:
+        line = line.strip()
+        if not line:
+            continue
+        
+        # 检查是否是domain行（没有逗号的行）
+        if ',' not in line:
+            current_domain = line
+        else:
+            # 这是attributes行
+            if current_domain:
+                attributes = [attr.strip() for attr in line.split(',')]
+                for attr in attributes:
+                    domain_attribute_pairs.append((current_domain, attr))
+    
+    return domain_attribute_pairs
+
+def load_templates(file_path):
+    """读取template_batch.json文件"""
+    with open(file_path, 'r', encoding='utf-8') as f:
+        templates = json.load(f)
+    return templates
+
+def batch_domain_attribute_pairs(pairs, batch_size=BATCH_SIZE):
+    """将domain-attribute pairs分批，每批batch_size个"""
+    batches = []
+    for i in range(0, len(pairs), batch_size):
+        batch = pairs[i:i + batch_size]
+        batches.append(batch)
+    return batches
+
+def generate_prompt(template, domain_attribute_pairs):
+    """生成prompt，将模板中的占位符替换为实际内容"""
+    # 格式化为: "domain1: attribute1, domain2: attribute2, ..."
+    pairs_text = ", ".join([f"{domain}: {attr}" for domain, attr in domain_attribute_pairs])
+    prompt = template.replace("{DOMAIN_ATTRIBUTE_PAIRS}", pairs_text)
+    
+    # 添加固定的布局要求
+    actual_count = len(domain_attribute_pairs)
+    layout_requirement = (
+        f" The output must be a single image with an exact 6:4 aspect ratio (landscape orientation, width greater than height). "
+        f"The image must contain exactly {actual_count} icons arranged in a strict grid of 6 columns (horizontal, left to right) and 4 rows (vertical, top to bottom). "
+        f"Do not rotate, transpose, or alter the grid orientation. "
+        f"No text, letters, numbers, labels, captions, or icon titles. Each icon must not include any titles or written elements. "
+        f"Use a pure white background only."
+    )
+    
+    prompt = prompt + layout_requirement
+    return prompt
+
+def generate_icons_for_batch(batch_idx, domain_attribute_pairs, templates, output_dir):
+    """为单个batch生成所有风格的icons"""
+    batch_dir = os.path.join(output_dir, f"batch_{batch_idx:04d}")
+    os.makedirs(batch_dir, exist_ok=True)
+    
+    print(f"\n{'='*80}")
+    print(f"批次 {batch_idx} (含 {len(domain_attribute_pairs)} 个 domain-attribute pairs)")
+    
+    success_count = 0
+    failed_count = 0
+    skipped_count = 0
+    
+    # 为每个style生成icons
+    for style_name, template in templates.items():
+        print(f"\n  风格: {style_name}")
+        
+        # 生成文件名
+        image_filename = f"batch_{batch_idx:04d}_{style_name}.png"
+        annotation_filename = f"batch_{batch_idx:04d}_{style_name}.txt"
+        
+        image_path = os.path.join(batch_dir, image_filename)
+        annotation_path = os.path.join(batch_dir, annotation_filename)
+        
+        # 断点续传：检查文件是否已存在
+        if os.path.exists(image_path) and os.path.exists(annotation_path):
+            print(f"    ⏭️  跳过（文件已存在）: {image_filename}")
+            skipped_count += 1
+            success_count += 1  # 已存在的文件计入成功数
+            continue
+        
+        try:
+            # 生成prompt
+            prompt = generate_prompt(template, domain_attribute_pairs)
+            
+            # 每个进程需要创建自己的API客户端
+            client = genai.Client(
+                api_key=API_KEY,
+                http_options={"base_url": "https://aihubmix.com/gemini"},
+            )
+            
+            # 调用API生成图像
+            response = client.models.generate_content(
+                model="gemini-3-pro-image-preview",
+                contents=prompt,
+                config=types.GenerateContentConfig(
+                    response_modalities=['TEXT', 'IMAGE'],
+                    image_config=types.ImageConfig(
+                        aspect_ratio=ASPECT_RATIO
+                    ),
+                ),
+            )
+            
+            # 保存图像和文本
+            for part in response.parts:
+                if part.text:
+                    print(f"    生成说明: {part.text[:100]}...")
+                elif image := part.as_image():
+                    image.save(image_path)
+                    print(f"    ✅ 图像已保存: {image_filename}")
+                    
+                    # 保存标注
+                    save_annotation(annotation_path, domain_attribute_pairs, style_name)
+                    print(f"    ✅ 标注已保存: {annotation_filename}")
+            
+            success_count += 1
+            
+        except Exception as e:
+            print(f"    ❌ 生成失败: {str(e)}")
+            failed_count += 1
+            continue
+    
+    if skipped_count > 0:
+        print(f"\n  批次 {batch_idx} 完成: ✅ {success_count} 成功 (含 {skipped_count} 个跳过), ❌ {failed_count} 失败")
+    else:
+        print(f"\n  批次 {batch_idx} 完成: ✅ {success_count} 成功, ❌ {failed_count} 失败")
+    return success_count, failed_count
+
+def save_annotation(file_path, domain_attribute_pairs, style):
+    """保存txt标注文件，格式：第一行style，后续每行domain, attribute"""
+    with open(file_path, 'w', encoding='utf-8') as f:
+        f.write(f"Style: {style}\n")
+        for domain, attr in domain_attribute_pairs:
+            f.write(f"{domain}, {attr}\n")
+
+def process_single_batch(args):
+    """处理单个batch的所有风格（用于并发处理）"""
+    batch_idx, batch, templates, output_dir = args
+    
+    # 跳过少于MIN_BATCH_SIZE的最后一批（如果不是第一批）
+    if len(batch) < MIN_BATCH_SIZE and batch_idx > 1:
+        print(f"\n⚠️ 跳过批次 {batch_idx}: 只有 {len(batch)} 个pairs (少于最小值 {MIN_BATCH_SIZE})")
+        return 0, 0, True  # success_count, failed_count, skipped
+    
+    success, failed = generate_icons_for_batch(batch_idx, batch, templates, output_dir)
+    return success, failed, False
+
+def main():
+    """主函数"""
+    print("="*80)
+    print("批量图标生成器 (扁平化模式)")
+    print("="*80)
+    
+    # 检查文件是否存在
+    domain_file = "domain_attributes.txt"
+    template_file = "template_batch.json"
+    
+    if not os.path.exists(domain_file):
+        print(f"❌ 错误: 找不到文件 {domain_file}")
+        return
+    
+    if not os.path.exists(template_file):
+        print(f"❌ 错误: 找不到文件 {template_file}")
+        return
+    
+    # 创建输出目录
+    os.makedirs(OUTPUT_DIR, exist_ok=True)
+    
+    # 加载数据
+    print(f"\n📖 加载domain-attribute pairs...")
+    domain_attribute_pairs = load_domain_attributes(domain_file)
+    print(f"   共加载 {len(domain_attribute_pairs)} 个 domain-attribute pairs")
+    
+    print(f"\n📖 加载模板...")
+    templates = load_templates(template_file)
+    print(f"   共加载 {len(templates)} 个风格模板: {', '.join(templates.keys())}")
+    
+    # 配置信息
+    print(f"\n⚙️  配置:")
+    print(f"   长宽比: {ASPECT_RATIO}")
+    print(f"   单批最大数量: {BATCH_SIZE}")
+    print(f"   单批最小数量: {MIN_BATCH_SIZE}")
+    print(f"   并发进程数: {NUM_PROCESSES}")
+    print(f"   输出目录: {OUTPUT_DIR}")
+    
+    # 分批
+    batches = batch_domain_attribute_pairs(domain_attribute_pairs, BATCH_SIZE)
+    print(f"\n📦 分成 {len(batches)} 个批次")
+    
+    # 扫描已存在的文件（断点续传预检）
+    print(f"\n🔍 扫描已存在的文件...")
+    existing_files = 0
+    total_expected_files = 0
+    for batch_idx, batch in enumerate(batches, 1):
+        if len(batch) < MIN_BATCH_SIZE and batch_idx > 1:
+            continue
+        batch_dir = os.path.join(OUTPUT_DIR, f"batch_{batch_idx:04d}")
+        for style_name in templates.keys():
+            total_expected_files += 1
+            image_filename = f"batch_{batch_idx:04d}_{style_name}.png"
+            annotation_filename = f"batch_{batch_idx:04d}_{style_name}.txt"
+            image_path = os.path.join(batch_dir, image_filename)
+            annotation_path = os.path.join(batch_dir, annotation_filename)
+            if os.path.exists(image_path) and os.path.exists(annotation_path):
+                existing_files += 1
+    
+    print(f"   已存在: {existing_files}/{total_expected_files} 个文件")
+    print(f"   需要生成: {total_expected_files - existing_files} 个文件")
+    
+    # 准备并发任务
+    tasks = []
+    for batch_idx, batch in enumerate(batches, 1):
+        tasks.append((batch_idx, batch, templates, OUTPUT_DIR))
+    
+    # 使用进程池并发处理多个batch
+    print(f"\n🚀 使用 {NUM_PROCESSES} 个进程并发处理批次...")
+    print(f"💡 提示: 已存在的文件将自动跳过（断点续传）")
+    total_success = 0
+    total_failed = 0
+    
+    with Pool(processes=NUM_PROCESSES) as pool:
+        results = pool.map(process_single_batch, tasks)
+    
+    # 统计结果
+    for success, failed, skipped in results:
+        if not skipped:
+            total_success += success
+            total_failed += failed
+    
+    print(f"\n{'='*80}")
+    print("✅ 全部生成完成！")
+    print(f"总计: ✅ {total_success} 成功, ❌ {total_failed} 失败")
+    print(f"输出目录: {OUTPUT_DIR}")
+    print("="*80)
+
+if __name__ == "__main__":
+    main()

icon_generation/count.py

@@ -0,0 +1,52 @@
+#!/usr/bin/env python3
+"""
+统计 generated_icons 目录下各个子目录中的 PNG 文件数量
+"""
+
+import os
+from pathlib import Path
+
+def count_png_files():
+    # 设置 generated_icons 目录路径
+    base_dir = Path(__file__).parent / "generated_icons"
+    
+    if not base_dir.exists():
+        print(f"错误: 目录 {base_dir} 不存在")
+        return
+    
+    # 存储每个子目录的统计结果
+    subdirs_stats = {}
+    total_png_count = 0
+    
+    # 遍历所有子目录
+    for subdir in sorted(base_dir.iterdir()):
+        if subdir.is_dir():
+            # 统计当前子目录下的 PNG 文件数量
+            png_files = list(subdir.glob("*.png"))
+            png_count = len(png_files)
+            
+            subdirs_stats[subdir.name] = png_count
+            total_png_count += png_count
+    
+    # 打印结果
+    print("=" * 50)
+    print(f"{'子目录':<20} {'PNG 文件数量':>15}")
+    print("=" * 50)
+    
+    for subdir_name, count in subdirs_stats.items():
+        print(f"{subdir_name:<20} {count:>15}")
+    
+    print("=" * 50)
+    print(f"{'总计':<20} {total_png_count:>15}")
+    print("=" * 50)
+    
+    # 额外统计信息
+    print(f"\n子目录总数: {len(subdirs_stats)}")
+    print(f"PNG 文件总数: {total_png_count}")
+    
+    if subdirs_stats:
+        avg_png_per_dir = total_png_count / len(subdirs_stats)
+        print(f"平均每个子目录的 PNG 数量: {avg_png_per_dir:.2f}")
+
+if __name__ == "__main__":
+    count_png_files()

icon_generation/domain_attributes.txt

@@ -0,0 +1,957 @@
+Football Club
+Real Madrid, FC Barcelona, Manchester United, Manchester City, Liverpool FC, Chelsea FC, Arsenal FC, Bayern Munich, Borussia Dortmund, Paris Saint-Germain, Juventus, AC Milan, Inter Milan, AS Roma, Napoli, Atletico Madrid, Ajax, PSV Eindhoven, Benfica, FC Porto, Sporting CP, Celtic FC, Rangers FC, Galatasaray, Fenerbahce, Boca Juniors, River Plate, Flamengo, Santos FC
+
+Basketball Club
+Los Angeles Lakers, Boston Celtics, Golden State Warriors, Chicago Bulls, Miami Heat, San Antonio Spurs, Brooklyn Nets, New York Knicks, Philadelphia 76ers, Toronto Raptors, Dallas Mavericks, Houston Rockets, Phoenix Suns, Denver Nuggets, Milwaukee Bucks, Cleveland Cavaliers, Detroit Pistons, Atlanta Hawks, Indiana Pacers, Orlando Magic, Utah Jazz, Oklahoma City Thunder, Portland Trail Blazers, Sacramento Kings, LA Clippers, Minnesota Timberwolves, Memphis Grizzlies, New Orleans Pelicans, Washington Wizards
+
+Education Level
+Primary Education, Secondary Education, Upper Secondary Education, Vocational Education, Technical Education, Associate Degree, Bachelor’s Degree, Master’s Degree, Doctoral Degree, Postdoctoral Education, Professional Degree, Teacher Training, Special Education, Adult Education, Continuing Education, Distance Learning, Online Education, Homeschooling, Early Childhood Education, Preschool, Kindergarten
+
+Currency
+US Dollar, Euro, British Pound, Japanese Yen, Chinese Yuan, Canadian Dollar, Australian Dollar, Swiss Franc, Indian Rupee, Russian Ruble, South Korean Won, Singapore Dollar, Hong Kong Dollar, New Zealand Dollar, Brazilian Real, Mexican Peso, Argentine Peso, Chilean Peso, Colombian Peso, South African Rand, Turkish Lira, Israeli Shekel, Saudi Riyal, UAE Dirham, Qatari Riyal, Kuwaiti Dinar, Bahraini Dinar, Omani Rial, Egyptian Pound, Nigerian Naira
+
+Landmark
+Eiffel Tower, Statue of Liberty, Great Wall of China, Taj Mahal, Colosseum, Machu Picchu, Petra, Christ the Redeemer, Big Ben, Sydney Opera House, Burj Khalifa, Golden Gate Bridge, Mount Fuji, Angkor Wat, Acropolis, Sagrada Familia, Alhambra, Stonehenge, Moai Statues, Chichen Itza, Mount Rushmore, Notre-Dame Cathedral, Louvre Museum, Buckingham Palace, Tower of London, Brandenburg Gate, Neuschwanstein Castle, Blue Mosque, Hagia Sophia, Grand Canyon
+
+Esports Game
+League of Legends, Dota 2, Counter-Strike 2, Valorant, Overwatch 2, Fortnite, PUBG: Battlegrounds, Apex Legends, Call of Duty, Rainbow Six Siege, StarCraft II, Warcraft III, Hearthstone, Rocket League, FIFA, EA Sports FC, Street Fighter, Tekken, Super Smash Bros., Smash Ultimate, Mobile Legends: Bang Bang, Arena of Valor, Free Fire, Clash Royale, Brawl Stars, Pokémon Unite, Teamfight Tactics, Age of Empires II, Halo Infinite
+
+Trend Direction
+Upward Trend, Downward Trend, Stable Trend, Volatile Trend, Fluctuating Trend, Seasonal Increase, Seasonal Decrease, Cyclical Trend, Exponential Growth, Linear Growth, Rapid Growth, Slow Growth, Plateau, Peak, Decline, Recovery, Rebound, Correction, Consolidation, Breakout
+
+Transportation Mode
+Car, Bus, Train (Rail), Subway (Metro), Bicycle, Walking, Taxi / Rideshare, Airplane (Aircraft), Ferry, Truck
+
+Smartphone Brand
+Apple, Samsung, Xiaomi, Oppo, Vivo, Google, Motorola, Huawei, OnePlus, Nokia
+
+Vehicle Type
+SUV, Sedan, Hatchback, Pickup Truck, Van, Bus, Motorcycle, Bicycle, Coupe, Semi-trailer Truck
+
+Energy Technologies
+Solar PV, Onshore wind, Hydropower, Nuclear power, Natural gas power, Lithium-ion battery, Hydrogen (production & storage), Grid transmission & distribution, Energy efficiency (LED lighting), Electric vehicles
+
+Energy Source
+Oil, Natural Gas, Coal, Hydropower, Solar, Wind, Nuclear, Biomass, Biofuel, Hydrogen, Geothermal, Biogas, Municipal Solid Waste, Tidal, Wave, Peat, Wood, Synthetic Fuels, Ocean Thermal Energy Conversion, Other Renewables
+
+Geographic Region
+Urban, Rural, Suburban, Coastal, Mountain, Plains, Forest, Desert, Island, Wetland
+
+Academic Subject
+Mathematics, Science, English, History, Biology, Chemistry, Physics, Computer Science, Economics, Psychology, Art, Music, Physical Education, Foreign Languages, Geography, Business, Engineering, Education, Philosophy, Political Science, Sociology, Statistics, Data Science, Information Technology, Environmental Science, Health Sciences, Finance, Accounting, Law, Architecture, Anthropology, Marketing, Media Studies, Religious Studies, Agriculture, Humanities
+
+Material
+Plastic, Wood, Steel, Concrete, Glass, Paper, Aluminum, Cotton, Rubber, Ceramic
+
+Product Brand
+Apple, Amazon, Google, Microsoft, Samsung, Nike, Coca-Cola, Toyota, Sony, McDonald's, Mercedes-Benz, BMW, Adidas, Ford, Honda, Louis Vuitton, Gucci, H&M, Zara, Uniqlo, Rolex, Tesla, LG, Intel, Dell, HP, Nintendo, Spotify, Skechers, New Balance, Puma, Converse, Reebok, ASICS, Under Armour, Tommy Hilfiger, Razer, Logitech, Corsair, SteelSeries, HyperX, Turtle Beach, Astro, Callaway, Brooks, Saucony, Carhartt, Diesel, Maserati, Fendi, Salomon, G-Star RAW, American Eagle
+
+Client Engagement Channel
+Website / Online Portal, Mobile App, Email, SMS / Text Messaging, Phone / Call Center, Live Chat / Chatbot, Branch / In-person Banking, Social Media, Push Notification, ATM / Kiosk, Financial Advisor, Video Conference, Mobile Web, Brokerage Platform, Robo-advisor / Digital Advisor, Partner / Channel Partner, API / Developer Integration, Direct Mail / Postal Mail, In-person Events / Seminars, Voice Assistant
+
+Park Type
+Urban Park, Playground, National Park, State Park, Botanical Garden, Zoo, Aquarium, Amusement Park, Nature Reserve, Campground
+
+Camera Form Factor
+Smartphone Camera, Mirrorless Camera, DSLR, Point-and-Shoot, Action Camera, Camcorder, Instant Camera, Film Camera, 360 Camera, Underwater/Waterproof Camera
+
+Residential Room Type
+Kitchen, Bathroom, Bedroom, Living Room, Dining Room, Garage, Laundry Room, Basement, Home Office, Closet
+
+Business Segment (Food & Beverage)
+Beverages, Snacks, Dairy, Bakery, Fresh Produce, Frozen Foods, Meat & Poultry, Prepared Foods, Confectionery, Refrigerated Foods, Plant-Based Foods, Food Ingredients & Commodities, Condiments & Sauces, Cereals & Grains, Seafood, Baby & Infant Nutrition, Pet Food, Alcoholic Beverages, Health & Functional Foods, Foodservice & Catering, Oils & Fats, Packaged & Shelf-stable Foods, Sugar & Sweeteners, Spices & Seasonings, Organic Foods, Ethnic & Specialty Foods, Retail Grocery, Logistics & Distribution, Packaging Materials & Solutions, Nutritionals & Supplements
+
+Leisure Activity
+Watching TV and Movies, Listening to Music, Socializing, Reading, Cooking, Fitness / Exercise, Video Gaming, Traveling, Gardening, Hiking
+
+Museum Object Type
+Ceramics, Textiles, Paintings, Sculptures, Prints and Drawings, Photographs, Furniture, Jewelry, Metalwork, Manuscripts, Books, Coins, Tools, Archaeological Materials, Weapons, Glass, Musical Instruments, Costume, Architectural Elements, Maps and Charts, Scientific Instruments, Ceremonial Objects, Inscribed Tablets, Ephemera, Time-based Media, Models and Replicas, Toys and Games, Medals and Tokens
+
+Pollution Source or Pollutant Type
+Particulate matter (PM2.5/PM10), Nitrogen oxides (NOx), Sulfur dioxide, Carbon monoxide, Ozone, Volatile organic compounds (VOCs), Plastics / microplastics, Agricultural runoff, Sewage / wastewater discharge, Industrial effluent (chemical waste), Pesticides, Nitrates, Phosphates, Urban runoff / stormwater, Oil spills, Heavy metals, Greenhouse gases (CO2, CH4), Black carbon (soot), Sediment (erosion/siltation), Toxic chemical spills, Radioactive contamination, Thermal pollution, Ammonia, Light pollution, Noise pollution
+
+Medical Injury Type
+Fracture, Sprain, Strain, Laceration, Burn, Concussion, Dislocation, Spinal cord injury, Traumatic brain injury, Amputation
+
+Dietary Preference or Restriction
+No Dietary Restrictions, Vegetarian, Gluten-Free, Dairy-Free, Halal, Vegan, Lactose-Free, Nut-Free, Kosher, Organic, Peanut-Free, Low-Carb, Keto, Low-Sodium, Pescatarian, Diabetic-Friendly, Soy-Free, Egg-Free, Shellfish-Free, Fish-Free, Sugar-Free, Paleo, Low-FODMAP
+
+Dietary Protein Source
+Chicken, Beef, Pork, Fish, Eggs, Milk, Beans, Tofu, Nuts, Shrimp
+
+Lighting Type
+LED, Fluorescent, Incandescent, Halogen, High-Intensity Discharge, Neon, Solar-powered, Smart Lighting, Fiber Optic
+
+Seafood Species
+Shrimp, Salmon, Tuna, Cod, Crab, Lobster, Oyster, Squid, Scallop, Mackerel
+
+Travel Category
+Flights, Hotels, Vacation Rentals, Car Rentals, Trains, Buses, Cruises, Tours & Activities, Travel Insurance, Travel Packages
+
+Cryptocurrency Name
+Bitcoin, Ethereum, Tether, USD Coin, XRP, Solana, Cardano, Dogecoin, Litecoin, Binance Coin
+
+Food Category
+Vegetables, Fruits, Grains, Dairy, Meat, Beverages, Snacks, Baked Goods, Seafood, Poultry, Eggs, Frozen Foods, Prepared Meals, Condiments & Sauces, Oils & Fats, Spices & Herbs, Canned & Preserved Foods, Sweets & Confectionery, Desserts, Legumes & Pulses, Nuts & Seeds, Breakfast Foods, Soups & Stews, Baby Food, Alcoholic Beverages, Plant-based Alternatives, Fast Food, Fermented Foods, Salads & Ready-to-eat Bowls, Cereals & Breakfast Grains
+
+Fashion Accessory Category
+Jewelry, Handbags, Wallets, Sunglasses, Watches, Hats, Scarves, Belts, Gloves, Hair Accessories
+
+Tourism Type
+Leisure tourism, Domestic tourism, International tourism, Business tourism, Cultural tourism, Adventure tourism, Ecotourism, Wellness tourism, Religious/pilgrimage tourism, Cruise tourism
+
+Interest Category
+Sports, Music, Movies, Television, Travel, Food & Cooking, Social Media, Fitness & Wellness, Gaming, Technology, Books & Reading, Fashion, Photography, Art, Arts & Crafts, Programming, Cars & Automotive, Science, History, Finance & Investing, Business & Entrepreneurship, Gardening, Pets & Animals, DIY & Home Improvement, Outdoor Activities, Board Games, Theater & Performing Arts, Dance, Languages, Comics & Graphic Novels, Collecting, Yoga & Meditation, Parenting & Family, Politics & Current Affairs, Tabletop Role-Playing Games, Education & Learning
+
+Financial Incentive Type
+Tax Credit, Tax Deduction, Rebate, Discount, Cashback, Grant, Subsidy, Voucher, Low-Interest Loan, Loan Guarantee
+
+Museum Focus
+Art, History, Natural History, Science, Children's Museum, Archaeology, Anthropology & Ethnography, Decorative Arts, Design, Contemporary Art, Photography, Technology & Industry, Maritime, Military, Historic House, Heritage, Religious, Music, Sports, Transportation, Aviation, Numismatics, Textiles, Fashion, Medical & Health, Space & Astronomy, Open-air / Living History, Specialty / Single-subject
+
+Waste Material Type
+Municipal solid waste, Food waste, Plastic, Paper, Yard waste, Cardboard, Construction and demolition waste, Glass, Metal, Industrial waste, Concrete, Electronic waste, Textiles, Used oil, Batteries, Sewage sludge, Hazardous waste, Paints and solvents, Rubber, Wood, Tires, Medical waste, Pharmaceutical waste
+
+Freight Cargo Type
+Containers, Crude oil, Refined petroleum products, Coal, Iron ore, Grain, Chemicals (liquid bulk), Automobiles (Ro-Ro), Refrigerated goods (Reefer cargo), Project cargo / Heavy lift / Oversize
+
+Funding Source
+Donations, Grants, Sponsorships, Ticket Sales, Membership Fees, Government Contracts, Loans, Product Sales, Service Fees, Investment Income
+
+Civil Engineering Structure Type
+Bridge, Tunnel, Road (Overpass/Underpass/Viaduct), Dam, Levee, Seawall, Canal, Retaining Wall, Pier/Wharf/Jetty, Reservoir
+
+Recipe Ingredient
+Salt, Olive Oil, Onion, Garlic, Eggs, Flour, Sugar, Milk, Butter, Chicken
+
+Software Application Name
+Google Chrome, Gmail, YouTube, WhatsApp, Facebook, Instagram, Microsoft Word, Microsoft Excel, Zoom, Google Drive
+
+Sofa Type
+Sectional, Sleeper sofa, Loveseat, Reclining sofa, Modular sofa, Futon, Chaise lounge, Chesterfield, Three-seater sofa, Settee
+
+Sales Offer Type
+Discount, Coupon / Promo Code, Free Shipping, Buy One Get One (BOGO), Bundle, Clearance Sale, Flash Sale, Free Trial, Cashback, Gift with Purchase
+
+Financial Product Type
+Checking Account, Savings Account, Credit Card, Debit Card, Mortgage, Personal Loan, Auto Loan, Brokerage Account, Certificate of Deposit (CD), Individual Retirement Account (IRA)
+
+Cuisine
+Italian, Chinese, Mexican, American, Indian, Japanese, Thai, Mediterranean, French, Middle Eastern
+
+Season (Calendar & Climatic)
+Spring, Summer, Autumn, Winter, Holiday season, Rainy/Wet season, Dry season, Hurricane/Typhoon/Cyclone season, Flu season, Tourist season
+
+Company Name
+Apple, Amazon, Google, Microsoft, Meta, Walmart, Tesla, Samsung, Toyota, Coca-Cola
+
+Live Performance Type
+Concert, Play, Musical, Stand-up Comedy, Dance, Ballet, Opera, Children's Theatre, Puppetry, Magic Show, Circus, Cabaret, Performance Art, Variety Show, Spoken Word / Poetry Slam, Improv Comedy, Physical Theatre, Orchestra / Symphony, Choral Performance, Recital, One-person Show (Monologue), Street Performance / Busking, Drag Show, Burlesque, Pantomime, Mime, Immersive / Interactive Theatre, Site-specific Performance, Multimedia / Digital Performance, Educational / School Performance
+
+Animal Species
+Dog, Cat, Chicken, Cattle, Horse, Pig, Elephant, Lion, Tiger, Bear
+
+Event Type
+Meeting, Conference, Workshop, Seminar, Concert, Festival, Sporting Event, Exhibition, Market/Fair, Parade, Protest/Rally, Wedding, Theater Performance, Film Screening, Dance Performance, Class/Training, Webinar/Virtual Event, Building Fire, Wildfire, Flood, Thunderstorm, Hailstorm, Heatwave, Drought, Tropical Cyclone, Tornado, Earthquake, Tsunami, Landslide, Sandstorm, Severe Winter Storm, Volcanic Eruption, Chemical Spill, Industrial Accident, Power Outage, Traffic Accident, Explosion, Mass Casualty Incident, Evacuation
+
+Mobile Phone Type
+Smartphone, Feature phone, Landline phone, VoIP phone, Flip phone, Rugged phone, Satellite phone, Desk phone
+
+Accommodation Type
+House, Apartment, Hotel, Vacation Rental, Condominium, Hostel, Bed and Breakfast, Resort, Motel, Cabin
+
+Operating System
+Android, iOS, Windows, macOS, Linux, Chrome OS
+
+Vehicle Powertrain Type
+Gasoline (ICE), Diesel (ICE), Battery Electric (BEV), Hybrid (HEV), Plug-in Hybrid (PHEV), Fuel Cell (FCEV), Natural Gas (CNG), Flex-Fuel (E85)
+
+Industry Sector
+Healthcare, Manufacturing, Technology, Finance, Retail, Construction, Education, Hospitality & Tourism, Real Estate, Transportation & Logistics, Energy, Agriculture, Automotive, Pharmaceuticals & Biotechnology, Food & Beverage, Consumer Goods (FMCG), Telecommunications, Media & Entertainment, Professional & Business Services, Government / Public Sector, Mining & Oil & Gas, Chemicals, Aerospace & Defense, Electronics & Electrical Equipment, Industrial Machinery & Equipment, Renewable Energy, Utilities, Software & SaaS, Consumer Electronics, Textiles & Apparel, Nonprofit / Social Services, Advertising & Marketing Services
+
+Web Browser
+Chrome, Safari, Edge, Firefox, Opera, Samsung Internet, UC Browser, Internet Explorer, Brave, Tor Browser
+
+Entertainment Genre
+Drama, Comedy, Action, Thriller, Crime, Romance, Romantic Comedy, Mystery, Horror, Documentary, Animation, Family, Adventure, Fantasy, Science Fiction, Superhero, Biography, History, Musical, War, Western, Reality, Sports, Coming-of-Age, Noir, Satire, Art House, Experimental
+
+Media Format
+Digital, Print, Video, Audio, Streaming, Television, Film, Podcast, Video Game, Live Performance
+
+Smart Home Device Category
+Smart Speakers & Voice Assistants, Smart Lighting, Smart Thermostats & Climate Control, Security Cameras, Video Doorbells, Smart Locks, Smart Plugs & Outlets, Smart Appliances, Smart Sensors (motion, contact, leak), Home Automation Hubs & Controllers
+
+Art Medium
+Painting, Drawing, Photography, Sculpture, Digital Art, Printmaking, Mixed Media, Ceramics, Textiles (Fiber Art), Watercolor, Oil Painting, Acrylic Painting, Pastel, Charcoal, Ink, Collage, Illustration, Glass, Metalwork, Woodwork, Encaustic, Fresco, Mosaic, Tapestry, Video Art, Film, Performance Art, Installation, Screenprint
+
+Product Category
+Electronics, Apparel, Food & Grocery, Home & Kitchen, Health & Personal Care, Books, Furniture, Toys & Games, Sports & Outdoors, Automotive
+
+Land Use Type
+Agriculture, Forest, Grassland, Wetland, Water, Residential, Commercial, Industrial, Transportation, Mining / Quarry
+
+Device Type
+Smartphone, Laptop, Desktop Computer, Tablet, Smart TV, Smartwatch, Gaming Console, Headphones, Printer, Router
+
+Crop Type
+Corn, Wheat, Rice, Soybean, Potato, Sugarcane, Cotton, Barley, Canola/Rapeseed, Peanut
+
+Coffee Brewing Method
+Drip Coffee, Espresso, Pour-over, French Press, Cold Brew, Instant Coffee, Pod Coffee, Moka Pot, Turkish Coffee
+
+Social Event Type
+Birthday Party, Wedding, Holiday Party, Dinner Party, Graduation Party, Baby Shower, Funeral / Memorial Service, House Party, Retirement Party
+
+Purchase Channel
+E-commerce Website, In-store (Retail), Mobile App, Third-party Marketplace, Click & Collect (Buy Online, Pick Up In Store), Subscription / Recurring Order, Telephone (Call Center / Phone Order), Wholesale / Distributor (B2B Channel), Social Commerce (Social Media Shops), Self-service Kiosk (In-store Kiosk)
+
+Investment Asset Class
+Equities, Fixed Income, Cash and Cash Equivalents, Real Estate, Commodities, Currencies (Foreign Exchange), Derivatives, Private Equity, Hedge Funds, Cryptocurrencies
+
+Expense Category
+Groceries, Rent, Utilities, Transportation, Restaurants & Dining, Fuel, Insurance, Healthcare & Medical, Taxes, Subscriptions & Streaming, Entertainment, Clothing & Apparel, Household Goods & Supplies, Home Maintenance & Repairs, Travel & Accommodation, Public Transport, Taxi & Ride-sharing, Parking & Tolls, Childcare & Education, Personal Care & Grooming, Gifts & Donations, Coffee & Snacks, Alcohol & Bars, Pet Care & Supplies, Professional Services, Office Supplies, Events & Catering, Venue Hire, Photography & Videography, Flowers & Decorations
+
+Music Genre
+Pop, Rock, Hip Hop, R&B, Electronic, Country, Latin, Classical, Jazz, Reggae
+
+Dwelling Type
+Single-Family Home, Apartment, Condominium, Townhouse, Multi-Family Home, Duplex, Manufactured Home, Studio Apartment, Cabin, Tiny House
+
+Time of Day
+Morning, Afternoon, Evening, Night, Dawn, Noon, Dusk, Midnight
+
+Produce (Fruit & Vegetable)
+Tomato, Potato, Onion, Banana, Apple, Lettuce, Carrot, Corn, Garlic, Cucumber, Bell pepper, Broccoli, Spinach, Avocado, Grapes, Orange, Strawberry, Mango, Lemon, Lime, Sweet potato, Cabbage, Cauliflower, Zucchini, Eggplant, Chili pepper, Pea, Green bean, Mushroom, Kale, Pumpkin, Pear, Peach, Pineapple, Watermelon, Melon, Beet, Radish, Celery, Asparagus, Basil, Parsley, Cilantro
+
+Medical Procedure
+Blood test, Vaccination, Colonoscopy, Appendectomy, Cesarean section, Vaginal delivery, Cataract surgery, Hip replacement, Knee replacement, Coronary angioplasty (PCI)
+
+Affectionate Gestures
+Hug, Kiss, Holding hands, Cuddling, Compliment, Love letter, Giving flowers, Romantic dinner, Surprise gift, Acts of service
+
+Mobile OS
+Android, iOS, HarmonyOS, Amazon Fire OS, KaiOS, Tizen, YunOS (AliOS), Series 40, Symbian, BlackBerry OS
+
+Medical Specialty
+Family Medicine, Internal Medicine, Pediatrics, Obstetrics and Gynecology, Emergency Medicine, Psychiatry, General Surgery, Cardiology, Neurology, Dermatology
+
+Health Topic
+Mental Health, Cardiovascular Disease, Cancer, Infectious Diseases, Diabetes, Respiratory Diseases, Nutrition, Physical Fitness, Women's Health, Substance Use Disorders
+
+Eating Occasion
+Breakfast, Lunch, Dinner, Snack, Brunch, Dessert, Late-night snack, On-the-go meal, Business meal, Holiday meal
+
+Storage Type
+Warehouse, Shipping container, Box, Pallet, Bin, Shelf, Refrigerator, Freezer, Tank, Cabinet
+
+Coffee Beverage / Preparation Style
+Espresso, Filter (Drip), Latte, Cappuccino, Americano, Iced Coffee, French Press, Cold Brew, Mocha, Instant
+
+Environmental Impact Categories
+Greenhouse Gas Emissions (CO2e), Energy Consumption, Water Use, Waste Generation, Air Pollution, Water Pollution, Land Use, Biodiversity Loss / Habitat Loss, Resource Depletion, Chemical Pollution / Toxicity
+
+Game Genre
+Action, Shooter, Role-Playing, Adventure, Puzzle, Sports, Strategy, Simulation, Racing, Horror
+
+Holiday Name
+Christmas Day, New Year's Day, New Year's Eve, Lunar New Year, Ramadan, Easter, Halloween, Valentine's Day, Eid al-Fitr, Diwali, Eid al-Adha, Hanukkah, Thanksgiving (United States), Mother's Day, Father's Day, Labor Day, St. Patrick's Day, Good Friday, Passover, Rosh Hashanah, Yom Kippur, Holi, Nowruz, Thanksgiving (Canada), Independence Day (United States), Memorial Day (United States), Veterans Day (United States), Canada Day, Bastille Day, Boxing Day, Cinco de Mayo, Kwanzaa
+
+Streaming Service
+Netflix, YouTube, Amazon Prime Video, Disney+, Hulu, Max, Apple TV+, Paramount+, Peacock, Pluto TV
+
+Weather Type
+Clear/Sunny, Cloudy/Partly Cloudy, Rain/Showers/Drizzle, Thunderstorm, Fog/Mist/Haze, Snow, Sleet/Freezing Rain, Windy, Hail, Dust/Sand
+
+Cooking Technique
+Baking, Boiling/Simmering, Frying, Grilling, Roasting, Steaming, Sautéing/Stir-frying, Braising/Stewing, Smoking, Pressure cooking
+
+Religious Affiliation
+Christianity, Islam, Hinduism, Buddhism, Judaism, Sikhism, Unaffiliated (Atheist/Agnostic/No religion), Indigenous/Traditional Religions, Other
+
+Sentiment
+Positive, Negative, Neutral, Mixed/Ambivalent, Support/Agree, Oppose/Disagree, Sarcastic, Factual, Unknown/Not Applicable
+
+Biome Type
+Tropical Rainforest, Temperate Forest, Boreal Forest (Taiga), Grassland/Savanna, Desert, Tundra, Mediterranean Shrubland (Chaparral), Wetlands (Marsh/Swamp/Peatland), Freshwater (Lakes/Rivers), Marine/Coastal (Reef/Estuary/Mangrove)
+
+Playground Feature
+Slide, Swing, Climbing structure (frame/wall/monkey bars), Seesaw, Sandbox, Roundabout/Merry-go-round, Playhouse, Balance elements (beam/stepping stones), Zip line, Trampoline
+
+Handicraft Category
+Sewing, Knitting, Crochet, Embroidery, Quilting, Jewelry making, Pottery/Ceramics, Woodworking, Leatherwork, Papercraft
+
+Retail Location Type
+Online / E-commerce, Downtown / Central Business District, Suburb, Main Street / High Street, Shopping Mall, Strip Mall / Strip Center, Outlet Center, Transit Hub / Airport Retail, Tourist / Entertainment District, Pop-up / Temporary Retail
+
+Art and Craft Techniques
+Painting, Drawing/Illustration, Photography, Sculpture, Printmaking, Ceramics, Digital Art, Textile Arts (weaving/embroidery), Woodworking/Carving, Jewelry/Metalworking
+
+Transportation Infrastructure Type
+Roads and Highways, Bridges, Railways, Airports, Ports (Seaports and Harbors), Public Transit (Buses, Trams, Metro), Tunnels, Bicycle Infrastructure (Bike Lanes, Cycle Paths), Pedestrian Infrastructure (Sidewalks, Footpaths), Intermodal Terminals (Freight)
+
+Physical Activity
+Walking, Running, Cycling, Swimming, Strength Training, Yoga, Team Sports, Dance, Hiking, Martial Arts
+
+Pet Category
+Dogs, Cats, Fish, Birds, Reptiles, Small Mammals, Amphibians
+
+Clothing Type
+Tops (T-shirt, Shirt, Blouse), Bottoms (Jeans, Pants, Shorts, Skirt, Leggings), Dresses, Outerwear (Jacket, Coat, Blazer), Sweaters (Sweater, Hoodie, Cardigan), Underwear, Socks, Shoes (Shoes, Boots, Sandals)
+
+Dance Discipline
+Ballet, Hip Hop, Contemporary, Jazz, Ballroom, Salsa, Tap, Breaking, Swing, Cultural/Street Styles (e.g., Afrobeats, K-Pop)
+
+Musical Instruments
+Piano, Guitar, Drums, Violin, Flute, Saxophone, Trumpet, Cello, Clarinet, Synthesizer
+
+Building Type
+Residential, Commercial, Office building, Retail, Mixed-use building, Industrial, Warehouse / Distribution, Hotel / Motel, Education (School/University), Healthcare (Hospital/Clinic)
+
+Household Appliance
+Refrigerator, Washing Machine, Dryer, Microwave, Oven / Cooktop, Dishwasher, Vacuum Cleaner, Television, Air Conditioner, Water Heater
+
+Incident Type
+Theft, Burglary, Robbery, Assault, Domestic Violence, Sexual Assault, Traffic Accident, Fraud, Vandalism, Drug Offense
+
+Occupation
+Healthcare (Nurse, Physician), Education (Teacher), Technology (Software Engineer), Business/Administration (Accountant, Administrative Assistant, Manager), Sales/Service (Retail Salesperson, Customer Service Representative, Cashier), Skilled Trades (Electrician, Plumber), Transportation/Logistics (Truck Driver), Construction (Construction Worker), Public Safety (Police Officer), Hospitality/Food (Cook)
+
+Milestone Type
+Birth, Graduation, Wedding, New Job, Promotion, Home Purchase, Retirement, Company Founded, Product Launch, Acquisition
+
+Cyberattack Type
+Phishing, Malware, Ransomware, Business Email Compromise (BEC), Distributed Denial of Service (DDoS), SQL Injection, Man-in-the-Middle (MITM), Credential Stuffing, Zero-Day Exploit, Insider Threat
+
+Art Subject
+Portrait, Landscape, Still Life, Abstract, Figurative, Cityscape, Seascape, Architecture, Floral, Wildlife, Historical, Religious, Mythological, Genre Scene, Self-Portrait, Nude, Conceptual, Fantasy, Social/Political, Allegory, Illustration, Sports, Industrial, Interior, Group Portrait, Narrative
+
+Food Item
+Bread, Rice, Potato, Chicken, Eggs, Milk, Pizza, Salad, Burger, Pasta, Sandwich, Soup, Fries, Fish, Noodles, Sushi, Curry, Tacos, Ice Cream, Cake, Pancake, Steak, Burrito, Dumpling, Cheese, Cereal, Chocolate, Hot Dog, Sausage, Shrimp, Beans, Oatmeal, Yogurt, Fruit, Vegetable
+
+Insect Common Name
+Mosquito, Housefly, Ant, Honey bee, Cockroach, Butterfly, Moth, Ladybug, Beetle, Fruit fly, Grasshopper, Locust, Dragonfly, Praying mantis, Termite, Aphid, Flea, Bed bug, Cicada, Firefly, Earwig, Weevil, Crane fly, Silkworm, Stink bug, Horsefly, Leafcutter ant, Soybean looper, Japanese beetle, Damselfly
+
+Insurance Type
+Health Insurance, Auto Insurance, Homeowners Insurance, Life Insurance, Renters Insurance, Travel Insurance, Pet Insurance, Disability Insurance, Dental Insurance, Vision Insurance, Long-Term Care Insurance, Umbrella Insurance, Motorcycle Insurance, Boat Insurance, Flood Insurance, Commercial Insurance, Workers' Compensation Insurance, Professional Liability Insurance, Cyber Insurance, Condo Insurance, Landlord Insurance, Title Insurance, Mortgage Insurance, RV Insurance, Mobile Home Insurance, Crop Insurance, Event Insurance, Identity Theft Insurance, Builder's Risk Insurance, Surety Bonds, Personal Articles Insurance
+
+Venue Seating Section
+General Admission, Standing Room, Floor, Orchestra, Balcony, Bleachers, Lower Bowl, Upper Bowl, Suite, Mezzanine
+
+Home Decor Category
+Wall Art, Rugs, Throw Pillows, Lighting, Curtains & Drapes, Indoor Plants, Mirrors, Candles, Vases, Throws & Blankets, Decorative Objects, Clocks, Shelving & Wall Storage, Baskets, Picture Frames, Sculptures, Planters & Pots, Tabletop Decor & Centerpieces, Decorative Trays, Bookends, Home Fragrance, Decorative Boxes, Accent Furniture, Mantel Decor, Seasonal Decor, Outdoor Decor, Table Linens, Wall Panels & Decals, Doormats & Welcome Mats, Storage & Organization
+
+Country / Organization
+United States, China, India, Russia, United Kingdom, France, Germany, Japan, South Korea, Canada, Australia, Brazil, Mexico, Argentina, Chile, Colombia, Peru, Spain, Portugal, Italy, Netherlands, Belgium, Switzerland, Austria, Sweden, Norway, Denmark, Finland, Poland, Czech Republic, Hungary, Romania, Bulgaria, Greece, Turkey, Israel, Saudi Arabia, United Arab Emirates, Qatar, Egypt, Morocco, South Africa, Nigeria, Kenya, Ethiopia, Ghana, Senegal, Tunisia, Algeria, Iran, United Nations, World Health Organization, World Bank, International Monetary Fund, World Trade Organization, European Union, African Union, ASEAN, NATO, OECD, UNICEF, UNESCO, UNHCR, International Red Cross, Amnesty International, Human Rights Watch, World Economic Forum, International Olympic Committee, FIFA, International Basketball Federation, Asian Development Bank, African Development Bank, European Central Bank, Federal Reserve, Asian Infrastructure Investment Bank, International Energy Agency, OPEC, International Atomic Energy Agency, Interpol, International Telecommunication Union, World Meteorological Organization, International Labour Organization, International Maritime Organization, World Intellectual Property Organization, International Civil Aviation Organization, International Organization for Migration, Doctors Without Borders, Save the Children, Greenpeace, World Wildlife Fund, International Chamber of Commerce, Transparency International, Global Fund, Gavi, International Crisis Group, Council of Europe, Arab League, Organization of American States, Commonwealth of Nations, World Customs Organization
+
+Computer Game
+Minecraft, Fortnite, League of Legends, Dota 2, Counter-Strike, Valorant, World of Warcraft, Grand Theft Auto V, The Witcher 3: Wild Hunt, Red Dead Redemption 2, Elden Ring, Dark Souls, Cyberpunk 2077, Call of Duty: Modern Warfare, Apex Legends, PUBG: Battlegrounds, Overwatch, StarCraft II, Diablo III, Hearthstone, The Elder Scrolls V: Skyrim, Fallout 4, Assassin’s Creed Valhalla, Assassin’s Creed Odyssey, Far Cry 5, Resident Evil 4, Monster Hunter: World, Final Fantasy XIV, The Legend of Zelda: Breath of the Wild, Super Mario Odyssey, Animal Crossing: New Horizons, Pokémon Red and Blue, Halo: Combat Evolved, Gears of War, God of War, Horizon Zero Dawn, Death Stranding, Metal Gear Solid V: The Phantom Pain, Sekiro: Shadows Die Twice, Civilization VI, Age of Empires II, The Sims 4, SimCity, Cities: Skylines, Kerbal Space Program, Terraria, Stardew Valley
+
+City
+New York City, Los Angeles, Chicago, Houston, Toronto, Vancouver, London, Paris, Berlin, Rome, Madrid, Barcelona, Amsterdam, Brussels, Zurich, Vienna, Stockholm, Copenhagen, Oslo, Helsinki, Tokyo, Osaka, Seoul, Beijing, Shanghai, Hong Kong, Singapore, Bangkok, Kuala Lumpur, Jakarta, Manila, Sydney, Melbourne, Auckland, Dubai, Abu Dhabi, Riyadh, Doha, Mumbai, Delhi, Bangalore, Chennai, Kolkata, São Paulo, Rio de Janeiro, Buenos Aires, Mexico City
+
+Driving Environment
+City, Highway, Suburban, Residential street, Intersection, Parking lot, Night driving, Wet road conditions, Construction zone, Snow or ice conditions
+
+Fitness Goal
+Weight Loss, General Fitness, Muscle Gain, Strength, Endurance, Flexibility, Mobility, Rehabilitation, Sports Performance, Weight Maintenance
+
+Communication Channel
+Website, Email, In-person, Phone call, Mobile app, SMS / Text message, Social media, Live chat, Messaging apps, Chatbot (automated chat), Push notification, Web form, Self-service portal, Knowledge base / FAQ, Video call, IVR / automated phone system, In-app messaging, Support ticketing system, Postal mail, Community forum, Kiosk, Voice assistant, Fax
+
+Grain Type
+Wheat, Rice, Corn, Barley, Oats, Rye, Sorghum, Millet, Quinoa, Buckwheat
+
+Furniture Category
+Chair, Table, Sofa, Bed, Desk, Cabinet, Dresser, Wardrobe, Bookcase, Nightstand
+
+Heritage Site Type
+Museum, Historic Building, Historic District, Archaeological Site, Monument, Religious Site, Castle, Palace, Memorial, Cultural Landscape
+
+Beverage Type
+Water, Coffee, Tea, Soda, Juice, Milk, Beer, Wine, Spirits, Energy Drink, Sparkling Water, Smoothie, Lemonade, Iced Tea, Cold Brew Coffee, Kombucha, Cocktail, Hot Chocolate, Plant-based Milk, Herbal Tea, Milkshake, Sports Drink, Cider, Kefir, Coconut Water, Espresso, Iced Coffee, Flavored Milk, Sake
+
+Research Publication Type
+Journal Article, Review Article, Preprint, Conference Paper, Book, Book Chapter, Thesis/Dissertation, Technical Report, Working Paper, Dataset, Letter / Short Communication, Conference Abstract, Editorial, Methods / Protocol, Patent, Case Report, Conference Poster, Monograph, Book Review, White Paper, Policy Brief, Software / Code Release, Commentary, Standard (Technical Standard), Erratum / Correction, Retraction
+
+Earring Type
+Studs, Hoops, Dangles, Clip-On Earrings, Ear Cuffs, Chandeliers, Huggies, Threaders
+
+Common Illicit and Recreational Drugs
+Alcohol, Nicotine, Caffeine, Cannabis, Cocaine, Methamphetamine, MDMA, Heroin, Benzodiazepines, Fentanyl
+
+Wearable Accessory Type
+Sunglasses, Eyeglasses, Belt, Bag, Hat, Scarf, Ring, Earrings, Necklace, Watch, Smartwatch, Fitness tracker, Bracelet, Gloves, Tie, Headband, Anklet, Cufflinks, Brooch, Hair accessory, Locket, Pocket square
+
+Play Activity Type
+Free Play, Imaginative/Role Play, Physical Play, Outdoor Play, Social/Cooperative Play, Construction/Building Play, Arts & Crafts, Ball Play, Board Games, Puzzle Play, Sensory Play, Water Play, Nature Exploration, Team/Organized Sports, Digital/Screen Play, Music and Movement, STEM Play, Manipulative/Fine Motor Play, Rough-and-Tumble Play, Independent/Solitary Play, Exploratory/Discovery Play, Card Games, Chasing/Tag Games, Creative Storytelling
+
+Music Distribution Format
+On-demand streaming, Terrestrial radio, Digital download, Physical media (CD/Vinyl/Cassette), Music video, Live performance, Short-form social media clips, Satellite radio, Ringtone, Sheet music
+
+Sports Court Surface
+Concrete, Asphalt, Hardwood, Grass, Artificial turf, Clay, Sand, Rubber, Carpet, Modular plastic tiles
+
+Application Category (App Store)
+Social Networking, Games, Entertainment, Productivity, Utilities, Finance, Health & Fitness, Shopping, Communication, Photo & Video, Music, Education, Travel, Food & Drink, News, Sports, Business, Navigation, Lifestyle, Weather, Books, Reference, Medical, Family, Events, Dating, Home Automation, Art & Design
+
+Exercise Type
+Walking, Running, Strength Training, Yoga, Cycling, Indoor Cycling, Weightlifting, High-Intensity Interval Training, Swimming, Pilates, CrossFit, Calisthenics, Zumba, Boxing, Kickboxing, Barre, Dance Fitness, Circuit Training, Rowing, Hiking, Aerobics, Mobility/Stretching, Martial Arts, Tai Chi, Functional Training, Climbing, Powerlifting, Olympic Weightlifting, Sprint Training, Stair Climbing
+
+Streaming Platforms
+Netflix, YouTube, Amazon Prime Video, Disney+, Hulu, Twitch, Spotify, Apple Music, Max, Paramount+
+
+Cultivation Environment
+Open Field, Backyard Garden, Container Gardening, Greenhouse, Plant Nursery, Raised Bed, Orchard, Hydroponics, Vertical Farm, Aquaponics
+
+Sensor Type
+Camera, Accelerometer, Microphone, Temperature Sensor, GNSS / GPS Receiver, Ambient Light Sensor, Capacitive Touch Sensor, Gyroscope, Magnetometer, Pressure Sensor, Humidity Sensor, Voltage Sensor, Current Sensor, Proximity Sensor, Ultrasonic Sensor, Radar, LiDAR, Thermal Imaging Camera, Vibration Sensor, Gas Sensor, CO2 Sensor, Particulate Matter (PM) Sensor, Force Sensor, Flow Sensor, Level Sensor, pH Sensor, Hall Effect Sensor, Inertial Measurement Unit (IMU)
+
+Wine Grape Varieties and Styles
+Cabernet Sauvignon, Chardonnay, Merlot, Pinot Noir, Sauvignon Blanc, Syrah/Shiraz, Riesling, Malbec, Rosé, Sparkling
+
+Manufacturing Step
+Material Procurement, Machining, Forming, Casting, Welding, Surface Treatment, Assembly, Quality Inspection, Testing, Packaging
+
+Amusement Ride Type
+Roller Coaster, Ferris Wheel, Carousel, Bumper Cars, Water Ride, Drop Tower, Dark Ride, Motion Simulator, Teacups, 4D Cinema
+
+Dish Type
+Pizza, Burger, Pasta, Salad, Sushi, Steak, Ramen, Taco, Sandwich, Soup, Fried Chicken, Curry, Noodles, Barbecue, Seafood, Dumplings, Wrap, Stir-fry, Kebab, Dim Sum, Paella, Risotto, Omelette, Pancake, Hot Pot, Sashimi, Tapas, Dessert
+
+Retailer Name
+Amazon, Walmart, Alibaba, eBay, Costco, Kroger, Target, The Home Depot, Walgreens, CVS Pharmacy, Tesco, Carrefour, IKEA, Lidl, Aldi, Best Buy, Macy's, Nordstrom, H&M, Zara, Sephora, Ulta Beauty, Dollar General, Dollar Tree, 7-Eleven, Sam's Club, JD.com, Flipkart, Mercado Libre, Coupang
+
+Pesticide Type (Target/Function)
+Herbicide, Insecticide, Fungicide, Rodenticide, Biopesticide, Acaricide, Nematicide, Fumigant, Repellent, Plant Growth Regulator
+
+Vehicle Maintenance Type
+Oil Change, Tire Rotation, Tire Replacement, Brake Inspection, Brake Pad Replacement, Battery Test & Replacement, Wheel Alignment, Fluid Check & Top-off, Engine Diagnostic (OBD Scan), A/C Service & Refrigerant Recharge
+
+Painting Medium
+Acrylic, Oil, Watercolor, Digital painting, Soft pastel, Gouache, Mixed media, Ink, Spray paint, Airbrush, Oil pastel, Egg tempera, Encaustic, Casein, Alcohol ink, Enamel, Charcoal, Collage, Metalpoint, Water‑mixable oil, Acrylic ink, Acrylic gouache, Marker, Oil stick, Gilding
+
+Craft Supplies
+Paper, Glue, Paint, Scissors, Yarn, Fabric, Markers, Pencils, Brushes, Canvas, Cardstock, Stickers, Adhesive tape, Hot glue gun, Glue sticks, Beads, Buttons, Ribbon, Wire, Clasps, Clay, Felt, Foam sheets, Pipe cleaners, Sequins, Glitter, Stencils, Stamps, Ink pads, Embroidery floss, Needles, Googly eyes, Craft knife, Cutting mat
+
+Project Phase
+Initiation, Planning, Design, Procurement, Construction, Testing, Commissioning, Handover, Operation, Closeout
+
+Dominant Forest Cover Type
+Tropical Rainforest, Temperate Rainforest, Deciduous Forest, Coniferous Forest, Boreal Forest (Taiga), Mixed Forest, Mangrove Forest, Woodland, Plantation Forest, Riparian Forest
+
+Festival Theme
+Music, Food, Art, Film, Cultural, Street Fair, Beer, Wine, Family, Parade
+
+Camera Lens Type
+Zoom lens, Prime lens, Wide-angle lens, Telephoto lens, Macro lens, Fisheye lens, Tilt-shift lens, Pancake lens
+
+Concession Stand Product
+Popcorn, Soda, Candy, Bottled Water, Beer, Nachos, Pretzel, Hot Dog, Pizza Slice, Ice Cream
+
+Climate Classification
+Tropical, Arid, Temperate, Continental, Mediterranean, Oceanic (Maritime), Subtropical, Polar
+
+Sports Facility Type
+Gym / Fitness Center, Soccer Field / Football Pitch, Basketball Court, Swimming Pool, Tennis Court, Running Track / Athletics Track, Baseball Field, Indoor Arena, Golf Course, Ice Rink
+
+Finishing Position
+First, Second, Third, Top 10, Finalist, Semi-finalist, Quarter-finalist, Did Not Finish, Disqualified, Tied
+
+Sustainability Category
+Climate Change, Renewable Energy, Energy Efficiency, Waste Management, Recycling, Water Management, Sustainable Agriculture, Sustainable Transportation, Biodiversity, Circular Economy
+
+Audio Listening Method
+Subscription streaming, Ad-supported streaming, Terrestrial radio (AM/FM), Podcasts, Internet radio (streamed radio stations), Local files (stored audio: MP3, FLAC), Purchased digital downloads, Physical CDs, Vinyl records, Casting / AirPlay / Bluetooth (device-to-speaker)
+
+Travel Market Segment
+Leisure, Business, Family, Budget, Luxury, Solo, Adventure, Romantic/Honeymoon, Cruise, Meetings & Events (MICE)
+
+Occasion (Usage Scenario)
+Everyday / Daily Wear, Work / Office, Casual Outing, Travel / Vacation, Party, Wedding, Formal Event, Job Interview, Holiday (e.g., Christmas/New Year's), Funeral
+
+Major Animal Groups
+Mammals, Birds, Fish, Reptiles, Amphibians, Insects, Arachnids, Mollusks, Crustaceans, Worms
+
+Alcoholic Beverage Type
+Beer, Wine, Whiskey, Vodka, Rum, Gin, Tequila, Champagne / Sparkling Wine, Cocktail, Sake
+
+Establishment Type
+Restaurant, Cafe / Coffee Shop, Bar / Pub, Grocery Store / Supermarket, Convenience Store, Retail Store, Hotel, Hospital / Clinic, School, Bank Branch
+
+Website Content Section
+News, Sports, Business, Technology, Entertainment, Lifestyle, Health, Politics, Travel, Opinion
+
+Property Flooring Type
+Hardwood, Carpet, Tile, Vinyl, Laminate, Concrete, Natural stone, Bamboo, Cork, Linoleum
+
+Vehicle Propulsion Type
+Gasoline (Petrol), Diesel, Battery Electric (BEV), Hybrid Electric (HEV), Plug-in Hybrid Electric (PHEV), CNG (Compressed Natural Gas), LPG (Liquefied Petroleum Gas), Fuel Cell Electric (Hydrogen), Ethanol / Flex-Fuel, Jet Engine / Turbine
+
+Retail Store Category
+Grocery, Convenience Store, Department Store, Discount Store, Pharmacy, Clothing Store, Electronics Store, Home Improvement Store, E-commerce / Online Retailer, Furniture Store
+
+Email Service Provider
+Gmail, Outlook.com, Yahoo Mail, iCloud Mail, ProtonMail, Yandex Mail, Zoho Mail, Custom/Corporate Email
+
+Costume Character
+Witch, Superhero, Princess, Vampire, Ghost, Zombie, Pirate, Clown, Wizard, Ninja
+
+Livestock Species
+Cattle, Pig, Chicken, Sheep, Goat, Aquaculture, Duck, Turkey, Rabbit, Bees
+
+Historical Period
+Prehistory, Ancient Egypt, Ancient Greece, Ancient Rome, Medieval Period, Renaissance, Industrial Revolution, World War I, World War II, Cold War
+
+Class Type
+Yoga, Pilates, Strength Training, High-Intensity Interval Training, Cardio, Spinning, Zumba, Cooking, Baking, Dance
+
+Religious Symbol
+Cross, Crescent and Star, Star of David, Om (Aum), Yin Yang, Dharma Wheel, Menorah, Hamsa, Ankh, Lotus
+
+Fruit Name
+Banana, Apple, Orange, Grape, Strawberry, Watermelon, Mango, Pineapple, Lemon, Pear, Blueberry, Cherry, Avocado, Grapefruit, Kiwi, Papaya, Plum, Peach, Apricot, Pomegranate, Raspberry, Coconut, Cantaloupe, Honeydew, Fig, Date, Lychee, Jackfruit, Passionfruit, Guava, Persimmon, Cranberry, Mandarin, Starfruit, Kumquat
+
+Pollution Source
+Transportation, Industrial Sources, Municipal Sewage and Wastewater, Agricultural Runoff, Stormwater and Urban Runoff, Landfills and Solid Waste Disposal, Residential Household Waste, Energy Production, Oil and Gas Operations, Plastic Waste and Marine Debris, Mining and Extractive Activities, Shipping and Ports, Accidental Chemical Spills, Hazardous Waste Sites, Pesticide and Fertilizer Use, Livestock Operations, Atmospheric Deposition, Construction and Demolition, Forestry and Logging, Fishing and Aquaculture, Industrial Air Emissions, Road Salt and De-icing, Illegal Dumping and Littering, Nonpoint Source Pollution, Point Source Discharges, Natural Sources, Tourism and Recreational Activities
+
+Dietary Food Group
+Vegetables, Fruits, Grains, Protein foods, Dairy, Fats and oils, Beverages, Sugars and sweets, Processed foods, Snacks
+
+Neighborhood Name
+Downtown, Midtown, Uptown, Suburb, Old Town, Historic District, Financial District, University District, Arts District, Waterfront
+
+Competition Gender Division
+Men, Women, Mixed, Open, Boys, Girls, Non-binary, Unspecified, Other
+
+Irrigation Source Type
+Groundwater, Surface water, Canal, Rainwater harvesting, Treated wastewater (recycled/effluent), Municipal / tap water, Desalinated water, Stormwater / surface runoff, Other / unspecified
+
+Gift Category
+Gift Cards, Clothing & Apparel, Electronics, Beauty & Personal Care, Home Decor, Books, Food & Gourmet, Toys & Games, Experiences (tickets, classes, travel), Flowers & Plants
+
+Vessel Type
+Container Ship, Bulk Carrier, Tanker, General Cargo Ship, Fishing Vessel, Ferry, Tug, Cruise Ship, Barge, Naval Vessel
+
+Online Video Platforms
+YouTube, TikTok, Netflix, Facebook Watch, Instagram, Amazon Prime Video, Disney+, Tencent Video, Twitch, Hulu, HBO Max, Apple TV+, iQIYI, Disney+ Hotstar, Snapchat, Bilibili, Youku, Tubi, Dailymotion, Vimeo, Peacock, Paramount+, Pluto TV, Roku Channel, Sling TV, FuboTV, ESPN+, DAZN, Crunchyroll, Rakuten Viki, Mubi, Shudder, CuriosityStream, Kanopy, Crackle, Viu
+
+Tropical Cyclone Intensity Category (Saffir–Simpson and related classifications)
+Tropical Depression, Tropical Storm, Category 1, Category 2, Category 3, Category 4, Category 5, Extratropical Cyclone, Post-Tropical Cyclone
+
+Internet Connectivity Status
+Online, Offline, Limited Connectivity, Intermittent Connectivity, Connecting, Connection Failed, Authentication Required, Captive Portal, Throttled, Maintenance
+
+Water Supply Type
+Piped household connection, Public tap / standpipe, Packaged / bottled water, Vendor-delivered water (tanker / cart), Borehole (drilled well), Dug well, Surface water (river / lake / pond / stream), Rainwater harvesting, Spring, Other / unspecified source
+
+Performance Terrain Type
+Road, Trail, Gravel Road, Dirt Road, Rocky Terrain, Sand / Beach, Mud, Snow / Ice, Grass, Mixed Terrain
+
+Software Application
+Web Browser, Email, Messaging/Chat, Social Media, Video Conferencing, Office Productivity, Cloud Storage, Streaming Media, Design/Photo Editing, Learning/Education
+
+Result Status (Test/Operation)
+Pass, Fail, Skipped, Not Run, Error, Pending, Running, Cancelled, Timed Out, Inconclusive
+
+Weightlifting and Strength Training Exercises
+Squat, Bench Press, Deadlift, Pull-up, Overhead Press, Row, Lunge, Plank, Bicep Curl, Kettlebell Swing
+
+Access Level
+No Access, Guest, Read-Only, Editor (Read/Write), Administrator, Owner
+
+Therapeutic Area
+Oncology, Cardiology, Infectious Diseases, Neurology, Psychiatry, Endocrinology, Respiratory, Gastroenterology, Dermatology, Rheumatology, Nephrology, Hematology, Pediatrics, Obstetrics & Gynecology, Allergy/Immunology, Pain Management, Ophthalmology, Otolaryngology, Geriatrics, Orthopedics, Emergency Medicine, Critical Care, Transplantation, Rare Diseases, Addiction Medicine, Genetics & Genomic Medicine, Public Health & Preventive Medicine, Vaccines, Sleep Medicine, Sports Medicine, Dentistry
+
+Employment Type
+Full-time, Part-time, Permanent, Temporary, Contract (Fixed-term), Independent Contractor, Internship, Seasonal, Remote, Hybrid
+
+Event Category
+Conference, Concert, Sports Event, Wedding, Trade Show, Exhibition, Workshop, Meetup, Webinar, Fundraiser
+
+Wellness Program Type
+Fitness Classes, Health Screenings, Nutrition Counseling, Mental Health Counseling, Employee Assistance Program, Smoking Cessation, Weight Management, Stress Management, Health Coaching, Wellness Challenges
+
+Packaging Material
+Plastic, Paper, Cardboard, Glass, Aluminum, Steel, Wood, Composite (Laminate), Bioplastic, Foam
+
+ADAS Feature
+Rear View Camera, Automatic Emergency Braking, Lane Departure Warning, Lane Keeping Assist, Adaptive Cruise Control, Blind Spot Monitoring, Parking Sensors, Forward Collision Warning, Pedestrian Detection, Traffic Sign Recognition
+
+Workout Modality
+Walking, Running, Strength Training, Yoga, Cycling, Swimming, HIIT, Pilates, Hiking, Martial Arts
+
+Common Pest Type
+Ants, Mosquitoes, Flies, Cockroaches, Rodents, Termites, Bed Bugs, Spiders, Ticks, Fleas
+
+Patrol Method
+Foot, Vehicle (Patrol car), Bicycle, Motorcycle, Mobile (roving) patrol, Static (Fixed-post / Checkpoint), Remote video surveillance (CCTV), Drone (UAV), K9 (Canine), Community / Neighborhood watch
+
+Sustainable Building Feature
+LED lighting, High-quality insulation, High-performance windows, Energy-efficient HVAC systems, Solar panels, Heat pumps, Smart energy management systems, Water-efficient fixtures, Low-VOC materials, EV charging stations
+
+Fatal Incident Type
+Motor vehicle collision, Fall, Suicide (intentional self-harm), Poisoning / drug overdose, Drowning, Homicide / assault, Structure fire, Industrial / workplace accident, Aviation accident, Electrocution
+
+Learning Resource Type
+Videos, Articles, Online Courses, Tutorials, E-books, Podcasts, Textbooks, Webinars, Documentation, Forums / Discussion Threads
+
+Service Category
+Electricity, Water, Internet & Telecommunications, Waste Management, Transportation, Healthcare Services, Emergency Services, Education Services, Sanitation, Natural Gas
+
+Manufacturing Process
+Assembly, Machining, Cutting, Welding, Injection Molding, Casting, Forging, Stamping, Extrusion, Additive Manufacturing
+
+Pest Control Technique
+Integrated Pest Management (IPM), Chemical control (pesticides), Biological control, Trapping, Exclusion / Barriers, Baiting, Monitoring / Surveillance, Habitat modification, Fumigation, Manual removal
+
+Personal Financial Concern Category
+Day-to-day living expenses, Housing affordability, Income loss / unemployment, High consumer debt (credit card debt), Unexpected medical bills, Retirement savings shortfall, Inflation and rising prices, Insufficient emergency fund, Student loan debt, Insurance costs (health, auto, home, life)
+
+Military Platform Type
+Fighter Aircraft, Transport Aircraft, Attack Helicopter, Main Battle Tank, Armored Personnel Carrier (APC), Destroyer, Frigate, Submarine, Aircraft Carrier, Self-Propelled Artillery
+
+Subscription Plan Tier
+Free, Free Trial, Basic, Standard, Premium, Business, Enterprise, Student, Family
+
+Organization Sector
+Healthcare & Medical, Education, Government / Public Sector, Retail & E-commerce, Finance & Banking, Information Technology (IT), Manufacturing, Construction, Real Estate, Transportation & Logistics, Professional Services, Nonprofit / NGO, Agriculture & Forestry, Hospitality & Tourism, Media & Entertainment, Energy & Utilities, Legal Services, Telecommunications, Insurance, Pharmaceuticals & Biotechnology, Arts & Culture, Environmental & Conservation, Religious / Faith-based, Sports & Recreation, Research & Development, Defense & Aerospace, Mining & Extraction, Consumer Goods / FMCG, Wholesale & Distribution
+
+Project Category
+New Construction, Renovation / Remodeling, Residential, Commercial, Infrastructure, Site Development, Demolition, Adaptive Reuse, Landscaping, Environmental Restoration
+
+Footwear Style
+Sneakers, Boots, Sandals, Heels, Flats, Loafers, Oxfords, Slippers, Flip-flops, Running Shoes
+
+Forage Type
+Pasture, Grass Hay, Alfalfa, Silage, Haylage, Straw, Rangeland, Clover, Crop Residues
+
+Clinical Service Type
+Primary Care Visit, Telehealth Visit, Preventive Care, Lab Services, Diagnostic Imaging, Medication Management, Emergency Department Visit, Surgical Consultation, Physical Therapy, Individual Therapy
+
+Hat Style
+Baseball cap, Beanie, Bucket hat, Sun hat, Fedora, Cowboy hat, Beret, Flat cap, Visor, Straw hat
+
+Technology Solution Type
+Sensors, Cloud Services, IoT Platforms, Edge Computing, AI/ML Solutions, Remote Monitoring, SCADA Systems, Energy Management Systems, Building Management Systems, EV Charging Stations
+
+Roofing Material/Type
+Asphalt Shingles, Metal Roofing, Clay Tile, Concrete Tile, Built-Up Roofing, EPDM Rubber Roofing, TPO Roofing, Modified Bitumen, PVC Roofing, Slate, Wood Shingles, Wood Shakes, Composite Shingles, Synthetic Slate, Solar Tiles, Spray Polyurethane Foam, Green Roof, Stone-Coated Metal, Standing Seam Metal, Corrugated Metal, Thatch
+
+Insect Group (common names)
+Ant, Bee, Beetle, Butterfly, Moth, Mosquito, Fly, Wasp, Cockroach, Termite
+
+Movie Theater Format
+Standard 2D, 3D, IMAX, Dolby Cinema, Premium Large Format, Drive-In, Art-house Cinema, 4DX, Outdoor/Open-Air Cinema, Luxury Seating
+
+Food Item Name
+Milk, Bread, Eggs, Bottled Water, Rice, Pasta, Chicken Breast, Ground Beef, Cheese, Butter, Yogurt, Frozen Vegetables, Potatoes, Apples, Bananas, Tomatoes, Onions, Lettuce, Cooking Oil, Sugar, Salt, Coffee, Soda, Orange Juice, Ketchup, Chocolate Chip Cookie, Ice Cream, Potato Chips, Hamburger Bun, Coleslaw, Iced Tea, Lemonade, Baked Beans, Ground Turkey, Pasta Salad, Watermelon, Cheddar, Mayonnaise, Flour
+
+Medical Treatment Type
+Medication, Vaccination, Preventive care, Surgery, Physical therapy, Psychotherapy, Lifestyle intervention, Radiation therapy, Chemotherapy, Immunotherapy, Palliative care, Emergency care, Medical device therapy, Transplantation, Dialysis, Blood transfusion, Hormone therapy, Occupational therapy, Speech therapy, Cognitive behavioral therapy, Behavioral intervention, Nutritional therapy, Dietary supplement, Complementary and alternative medicine, Acupuncture, Hospice care, Pain management, Gene therapy, Stem cell therapy, Placebo
+
+Handbag Style
+Tote, Shoulder bag, Crossbody bag, Backpack, Clutch, Messenger bag, Satchel, Hobo bag, Duffel bag, Briefcase, Belt bag, Wallet, Wristlet, Bucket bag, Top-handle bag, Saddle bag, Sling bag, Drawstring bag, Cosmetic bag, Weekender bag, Evening bag, Minaudière, Bowling bag, Doctor bag, Camera bag, Laptop bag, Frame bag, Carryall
+
+Educational Program Type
+Undergraduate Degree Program, Graduate Degree Program, Certificate Program, Online Program, Professional Development, Internship, Apprenticeship, Study Abroad, Fellowship, Continuing Education
+
+Art Installation Type
+Sculpture, Public Art, Installation Art, Monument, Memorial, Site-specific Installation, Light Installation, Interactive Installation, Video Installation, Sound Installation
+
+Tour Activity
+Sightseeing tour, Walking tour, City tour, Cultural/heritage tour, Food tour, Hiking, Boat cruise, Cycling tour, Wildlife safari, Snorkeling
+
+Server Role
+Web Server, Database Server, Application Server, Load Balancer, File Server, Cache Server, Mail Server, DNS Server, Authentication / Directory Server, Proxy Server, Backup Server, Monitoring Server, Logging Server, CI/CD Server, Storage Server, VPN Server, FTP / SFTP Server, DHCP Server, NTP (Time) Server, Container Registry, Orchestration / Cluster Management Server, API Gateway, Edge / CDN Node, Print Server, Media Streaming Server, Certificate Authority (CA) Server, Analytics / Big Data Server, Game Server, Remote Desktop / Terminal Server, License Server
+
+Emotion Type
+Happiness, Sadness, Anger, Fear, Surprise, Disgust, Love, Anxiety, Trust, Anticipation, Guilt, Shame, Pride, Jealousy, Envy, Contempt, Boredom, Excitement, Relief, Awe, Hope, Grief, Nostalgia, Compassion, Embarrassment, Curiosity, Contentment, Frustration, Serenity, Loneliness, Desire, Interest
+
+Cause of Damage
+Water damage, Fire, Storm and wind damage, Theft, Plumbing failure (burst pipes/leaks), Flood (natural flooding), Impact / collision (vehicle, falling objects), Hail, Vandalism / malicious damage, Mold and mildew, Freeze / frost damage, Electrical damage / power surge, Explosion, Structural failure / collapse, Subsidence / ground movement, Pest infestation (e.g., termites, rodents), Corrosion / rust, Snow / ice load damage, Accidental damage (drops, spills, human error), Wear and tear / gradual deterioration, Chemical contamination / spill, Oil or fuel leak, Biological contamination (sewage, biohazard), Lightning strike, Heat / thermal damage (non-fire), Landslide / mudslide, Storm surge / coastal inundation, Glass breakage
+
+Healthcare Facility Type
+Primary Care Clinic, Pharmacy, Urgent Care Clinic, Community Health Center, General Hospital, Emergency Department, Outpatient Clinic, Ambulatory Surgery Center, Skilled Nursing Facility, Assisted Living Facility, Home Health Agency, Hospice, Rehabilitation Hospital, Behavioral Health Center, Diagnostic Laboratory, Imaging/ Radiology Center, Dialysis Center, Birthing Center, Children's Hospital, Teaching Hospital, Community Hospital, Critical Access Hospital, Long-Term Acute Care Hospital (LTACH), Trauma Center, VA Hospital, Mobile Clinic, Retail Clinic, Dental Clinic, Optometry / Eye Clinic, Blood Bank / Transfusion Center, Public Health Department, Sexual and Reproductive Health Clinic, Occupational Health Clinic, School Health Center, Telehealth Service, Outpatient Rehabilitation Center, Psychiatric Hospital
+
+Application Functionality
+Authentication / Login, Push Notifications, User Profile (Account Management), Payments / Checkout, Search, Maps & Location (GPS), Camera / Photo Capture, Offline Mode & Data Sync, Analytics & Reporting, In-app Messaging / Chat, Social Sharing, Security & Encryption, Image Recognition, Barcode / QR Scanning, Forms & Surveys, Scheduling / Calendar / Booking, File Upload / Download (Document Viewer), API Integration / Webhooks, E-commerce / Product Catalog, Video Conferencing / Live Streaming, Machine Learning Recommendations, Role-based Access Control (Permissions), Voice Calling / VoIP, Alerts & Alarms, Backup & Restore / Data Export, Customer Support / Helpdesk / Live Chat, Feedback, Ratings & Reviews, Inventory & Order Management, Gamification & Badges, Remote Monitoring & Telemetry, Augmented Reality (AR), Equipment Maintenance, Pest Identification
+
+Ecosystem / Habitat Type
+Forest, Grassland, Desert, Wetland, Freshwater, Marine, Agricultural land, Urban, Coastal
+
+Craft Materials
+Fabric, Paper, Yarn, Glue, Paint, Wood, Metal, Plastic/Resin, Leather, Wire
+
+Subsystem / Component Type
+Processor / Microcontroller, Sensor, Controller / Electronic Control Unit (ECU), Battery, Power Supply, Electric Motor, Inverter, Switch, Pump, Valve
+
+Flavor Profile (Tasting Descriptors)
+Sweet, Salty, Sour/Acidic, Bitter, Umami, Fruity, Spicy, Floral, Smoky, Nutty
+
+Transport Route Type
+Highway / Expressway, Local Road / Street, Arterial Road, Collector Road, Ramp (On/Off Ramp), Rail Line, Subway / Metro Line, Bus Route, Bicycle Route, Ferry Route
+
+Basic Human Needs
+Food and Nutrition, Clean Water, Housing, Healthcare, Sanitation, Education, Income and Livelihood, Safety and Security, Energy and Utilities, Transportation
+
+Fashion Style
+Casual, Formal, Business Casual, Streetwear, Athleisure, Vintage/Retro, Bohemian, Minimalist, Punk, Gothic
+
+Humanitarian Aid Sector
+Health, Food Security and Agriculture, Water, Sanitation and Hygiene (WASH), Shelter and Non-Food Items (NFI), Protection, Education, Nutrition, Cash and Voucher Assistance (CVA), Logistics, Coordination and Information Management
+
+Character Type
+Human, Animal, Robot/Android, Alien, Monster, Mythical Creature, Undead, Deity/Demon, Ghost/Spirit, Cyborg
+
+Aircraft Category (type/market segment)
+Airplane (fixed-wing), Helicopter (rotorcraft), Narrow-body airliner, Wide-body airliner, Business jet, Regional aircraft, Cargo aircraft (freighter), Military aircraft, Glider (sailplane), VTOL / eVTOL
+
+Benefit Recipient Type
+Children, Seniors, People with disabilities, Low-income households, Unemployed individuals, Refugees and asylum seekers, Homeless individuals, Veterans, Students, Indigenous peoples
+
+Forms of Folklore
+Myth, Legend, Folktale, Fairy tale, Fable, Proverb, Riddle, Folk song, Urban legend, Superstition
+
+Generational Cohort
+Generation Alpha, Generation Z, Millennials, Generation X, Baby Boomers, Silent Generation, Greatest Generation
+
+Vehicle Part Type
+Engine, Transmission, Brakes, Tires, Battery, Suspension, Steering, Electrical System, Fuel System, Exhaust System
+
+Ad Format
+Search Ads, Social Media Ads, Display (Banner) Ads, Video Ads, Email Marketing, In-App Ads, Native Ads, Connected TV (CTV) Ads, Audio Ads, Influencer Marketing
+
+Consumer Electronics & Computer Hardware Category
+Smartphones, Headphones & Earbuds, Laptops, Televisions, Tablets, Desktop Computers, Monitors, Smartwatches, Smart Speakers, Game Consoles, Printers & Scanners, Keyboards, Computer Mice, External Storage Devices, Storage Drives (HDDs & SSDs), Networking Equipment (Routers, Modems, Switches), Graphics Cards, CPUs (Processors), Digital Cameras, Streaming Media Players, Smart Home Devices, Security Cameras & Surveillance, Projectors, Power Banks, Network Attached Storage (NAS), Drones, Virtual Reality & Augmented Reality Headsets, Fitness Trackers, E-Readers, Motherboards
+
+U.S. Military Service Branch
+Army, Navy, Air Force, Marine Corps, Space Force, Coast Guard, National Guard, Merchant Marine
+
+App Permission
+Internet / Network Access, Location, Camera, Microphone, Storage / Files & Media, Contacts, Notifications / Push, Bluetooth, SMS, Biometric authentication (FaceID/TouchID)
+
+News Organization
+Associated Press, Reuters, BBC News, CNN, The New York Times, The Washington Post, The Wall Street Journal, Fox News, Al Jazeera, Bloomberg
+
+Construction & Infrastructure Project Type
+Residential, Commercial, Industrial, Roads & Highways, Bridges, Rail & Public Transit, Airports, Ports & Harbors, Water & Wastewater, Power & Energy
+
+Home Feature
+Garage, Backyard, Air Conditioning, Heating, Modern Kitchen, Laundry Room, Basement, Fireplace, Swimming Pool, Home Office
+
+Production Method
+Handmade, Machine-made, Locally made, Imported, Recycled, Second-hand / Vintage, Organic, Fair trade, Custom-made, 3D printed
+
+Pricing Basis
+Fixed, Hourly, Subscription, Usage-based, Per user, Per transaction, Commission, Revenue share, Milestone-based, Retainer
+
+Booking Channel
+Direct Website, Mobile App, Online Travel Agency, Phone Call, Walk-in / In-person, Travel Agent, Global Distribution System (GDS), API / B2B Integration, Social Media Booking, Third-party Reseller
+
+Educational Focus Area
+Early Childhood Education, Primary Education, Secondary Education, Higher Education, STEM Education, Literacy, Teacher Training, Vocational Education and Training, Special Education, Adult Education
+
+Cultural Institution Type
+Museum, Public Library, Art Gallery, Theater, Cultural Center, Historic Site, Botanical Garden, Zoo, Aquarium, Science Center, Archive, Planetarium, Children's Museum, Concert Hall, Opera House, Memorial, Monument, Historic House, Visitor Center, Cultural Institute, Performing Arts Center, Community Arts Center, Living History Museum, Archaeological Site, Place of Worship, Historic District, Ethnographic Museum, Exhibition Space, Research Institute, Heritage Center, Film Archive
+
+Medical Interventions
+Medication therapy, Vaccination, Lifestyle modification, Preventive screening, Surgery, Physical therapy, Psychological therapy, Chemotherapy, Radiation therapy, Dialysis
+
+Loyalty Program Feature
+Points, Sign-up Bonus, Tiered Rewards, Cashback, Referral Program, Personalized Offers, Free Shipping, Flexible Redemption Options, Exclusive Access, Priority Customer Service
+
+Racing Circuit Type
+Permanent road course, Street circuit, Oval, Drag strip, Kart circuit, Rally stage, Hill climb, Autocross course
+
+Competitor Brand (Fashion & Apparel)
+Zara, H&M, Uniqlo, Shein, Nike, Adidas, Gap, Primark, ASOS, Mango
+
+Hazard Type
+Fire / Flammability, Slip, Trip & Fall, Electrical Hazard, Chemical Hazard (toxic, irritant, corrosive), Biological / Infectious Hazard, Mechanical Hazard (cuts, crush, entanglement), Thermal Hazard (burns, scalds, extreme cold), Choking, Noise / Hearing Damage, Ergonomic Hazard (strain, repetitive motion)
+
+Pet Service Type
+Veterinary Care, Emergency Veterinary Care, Pet Grooming, Dog Walking, Pet Sitting, Pet Boarding, Dog Daycare, Pet Training, Pet Transportation, Pet Food & Supply Delivery
+
+Pet Wellness Package Type
+Wellness/Preventive Care, Vaccination, Puppy/Kitten, Adult Pet Package, Senior, Spay/Neuter, Dental Care, Diagnostic Testing, Flea and Tick Prevention, Heartworm Prevention
+
+Freight Transport Mode
+Trucking, Ocean Freight, Rail Freight, Air Freight, Pipeline Transport, Intermodal Transport, Inland Waterway (Barge), Courier / Parcel, Last-mile Delivery
+
+Personal Data Type
+Full Name, Email Address, Phone Number, Home Address, Date of Birth, National Identification Number, Passport Number, Driver's License Number, Credit Card Number, Bank Account Number
+
+Financial Transaction Type
+Payment, Transfer, Deposit, Withdrawal, Purchase, Bill Payment, Refund, Fee, Chargeback, Currency Exchange
+
+Spending Occasion
+Groceries, Bills & utilities, Rent / Mortgage, Transportation / Commuting, Dining out / Takeout, Healthcare & medical expenses, Education / Tuition payments, Clothing, Vacation travel, Gifts
+
+Home Improvement Project Type
+Interior Painting, Exterior Painting, Kitchen Remodel, Bathroom Remodel, Flooring Installation, Roofing, Window Replacement, Door Replacement, Plumbing Repair, Electrical Upgrade, HVAC Upgrade, Siding Replacement, Insulation, Basement Finishing, Deck Construction, Fence Installation, Garage Door Replacement, Driveway Paving, Gutter Installation, Landscaping, Mold Remediation, Waterproofing, Solar Panel Installation, Home Addition, Smart Home Installation, Appliance Installation, Accessibility Modifications, Chimney Repair, Sewer & Septic Repair, Attic Conversion, Pool Installation/Repair
+
+Ticket Category
+General Admission, Reserved Seating, Standing, VIP, Premium, Early Bird, Student, Child, Senior, Accessible
+
+Hardware Component Type
+Resistor, Capacitor, Integrated Circuit, Transistor, Diode, Connector, PCB, Cable, Battery, Power Supply, Processor (CPU), Microcontroller, Memory (RAM), Storage, Display, Sensor, Actuator, Motor, Switch, LED, Oscillator, Heatsink, Fan, Antenna, Camera Module, Speaker, Microphone, Transformer, Relay, FPGA, ASIC, Power Management IC, Enclosure, Controller, Lens
+
+Manufacturing Machine Type
+CNC Machine, Lathe, Milling Machine, Drill Press, Welding Machine, Injection Molding Machine, Laser Cutter, Industrial 3D Printer, Press Brake, Grinder
+
+Motorcycle Type
+Scooter, Standard (Naked), Cruiser, Sportbike, Touring, Adventure (ADV), Dual-sport, Dirt Bike (Off-road), Electric Motorcycle, Moped
+
+Light Color (illumination)
+Warm White, Cool White, Daylight, Red, Green, Blue, Yellow, Purple, Pink, Multicolor
+
+Endorsement Category (Products & Services)
+Apparel, Beverages, Footwear, Consumer Electronics, Automotive, Beauty & Personal Care, Watches & Jewelry, Food & Snacks, Sports Equipment, Financial Services, Health & Wellness (Supplements & Fitness), Eyewear, Accessories (Bags, Hats, etc.), Home Goods & Appliances, Travel & Hospitality, Telecommunications & Internet Services, Media & Streaming Services, Gaming & Esports, Alcoholic Beverages, Fragrance & Perfume, Children's & Baby Products, Household & Cleaning Products, Pet Products, Education & Online Learning, Real Estate & Property Services, Non-profit & Cause Campaigns, Sports Teams & Leagues, Luxury Goods, Professional Services, Government & Public Service Campaigns
+
+Communication Purpose
+Informational, Notification, Reminder, Transactional, Confirmation, Promotional, Personal Connection, Support, Request, Call to Action, Alert, Welcome / Onboarding, Update / Newsletter, Invitation, Follow-up, Feedback / Survey, Instruction / How-to, Announcement, Status Update, Persuasion / Advocacy, Fundraising, Civic Duty, Candidate Support, Community Engagement, Appreciation / Thank You, Apology, Complaint, Recruitment / Hiring, Farewell / Offboarding
+
+Motor Vehicle Collision Type
+Rear-end collision, Sideswipe collision, Side-impact (T-bone) collision, Backing/reverse collision, Head-on collision, Run-off-road / single-vehicle collision, Rollover collision, Collision with fixed object, Intersection collision, Parked-vehicle / parking-lot collision, Chain-reaction / multi-vehicle collision, Pedestrian collision, Bicycle (pedalcyclist) collision, Hit-and-run collision, Animal strike (wildlife/large animal) collision, Underride / override (truck underride) collision, Collision with guardrail or barrier, Truck- or bus-involved collision, Collision with road debris, Low-speed / minor parking impact, Mechanical-failure-related collision, Other / unspecified collision type
+
+Cultural Offering Type
+Streaming Music, Streaming Films, Cinemas (Theatrical Films), Concerts, Museums, Art Galleries, Festivals, Libraries, Theater (Performing Arts), Exhibitions, Heritage Sites, Cultural Centers, Dance Performances, Opera, Ballet, Film Festivals, Comedy Shows, Literary Events, Workshops and Classes, Street Art and Public Art, Cultural Tours, Monuments and Memorials, Religious Sites and Pilgrimages, Craft Fairs and Markets, Archives and Special Collections, Virtual Exhibitions and Online Cultural Programs, Community Arts Programs, Public Lectures, Folk Events and Traditional Celebrations
+
+Donor Classification
+Individual Donor, Corporate Donor, Foundation Donor, Institutional Donor, Major Donor, Recurring Donor, First-time Donor, Lapsed Donor, Anonymous Donor, In-Kind Donor
+
+Body Shape (Figure Type)
+Rectangle, Pear, Hourglass, Apple, Inverted Triangle, Athletic, Diamond, Spoon, Oval
+
+Personal Protective Equipment (PPE) Type
+Gloves, Safety glasses/goggles, Respirator, Hard hat, Hearing protection, High-visibility clothing, Safety boots, Face shield, Safety harness, Protective clothing (coveralls/lab coat)
+
+Academic Subject Category
+Mathematics, English Language Arts, Science, Social Studies/History, Foreign Languages, Computer Science, Arts, Physical Education, Economics/Business, Engineering
+
+Microphone Type
+Dynamic, Condenser, Ribbon, Lavalier, Shotgun, USB microphone, Wireless microphone, Headset microphone, Boundary (PZM), Handheld
+
+Pollinator Types
+Bees, Butterflies, Moths, Flies, Beetles, Wasps, Ants, Birds (nectar-feeding), Bats (nectar-feeding)
+
+Computer Peripherals
+Monitor, Keyboard, Mouse, Printer, Speakers/Headphones, Webcam, Microphone, External storage (HDD/SSD/USB drive), Docking station/USB hub, Scanner
+
+Agricultural Activities
+Planting/Seeding, Irrigation, Fertilization, Pest management, Weeding, Tillage/Land preparation, Harvesting, Crop rotation, Livestock feeding/grazing, Soil testing
+
+Mission Phase
+Mission Planning, Pre-flight, Launch/Ascent, In-flight/Cruise, On-orbit/Operations, Docking/Undocking, Re-entry, Landing, Recovery/Post-flight, Abort/Contingency
+
+Certification Level
+Basic, Intermediate, Advanced, Professional, Expert, Associate, Master, Certified, Accredited, Gold
+
+Leisure Amenities
+Swimming Pool, Fitness Center, Restaurant, Bar/Lounge, Spa, Sauna/Steam Room/Hot Tub, Beach Access, Golf Course, Kids Club/Playground, Ski Facilities
+
+Personal Relationship Type
+Family, Friend, Spouse/Partner, Parent, Child, Sibling, Colleague, Neighbor, Acquaintance, Guardian
+
+Outdoor Lighting Fixture Type
+Flood light, Wall pack, Post-top, Bollard, Path light, Canopy light, Accent / spot light, Sconce, Pendant, Pole-mounted area light
+
+Material Sourcing Type
+Cotton, Organic Cotton, Recycled Cotton, Polyester, Recycled Polyester, Nylon, Recycled Nylon, Wool, Leather, Linen (Flax)
+
+Driving Scenario
+Urban Driving, Highway Driving, Parking, Lane Change, Merging, Signalized Intersection, Pedestrian Crossing, Night Driving, Rain / Wet Road Conditions, Snow / Icy Conditions
+
+Skilled Trade
+Electrician, Plumber, Carpenter, HVAC Technician, Welder, Automotive Technician, Painter, Roofer, Flooring Installer, Mason, Glazier, Drywall Installer, Tile Setter, Cabinetmaker, Locksmith, Landscaper, Heavy Equipment Operator, Concrete Finisher, Sheet Metal Worker, Pipefitter, Boilermaker, CNC Machinist, Millwright, Insulation Installer, Crane Operator, Upholsterer, Barber, Chef, Diesel Mechanic, Appliance Repair Technician, Pest Control Technician
+
+Land Use / Development Type
+Residential, Commercial, Industrial, Mixed-use, Agricultural / Farming, Parks and Recreation, Transportation / Transit Infrastructure, Logistics / Warehouse / Distribution, Institutional, Open Space / Conservation
+
+Ritual Type
+Prayer, Holiday Observance, Marriage Ceremony, Funeral, Burial, Cremation, Coming-of-Age Ceremony, Pilgrimage, Meditation, Memorial Service
+
+High-Risk Patient Groups
+Older adults (65+), People with diabetes, People with cardiovascular disease, People with chronic respiratory disease, Immunocompromised individuals, Pregnant women, People with active cancer, People with chronic kidney disease, Infants (under 1 year), Residents of long-term care facilities
+
+Subject Area (Topic)
+Politics & Government, Technology, Health, Business, Science, Education, Entertainment, Sports, Environment & Climate, Finance, Economy, Arts & Culture, Travel, History, Law & Public Policy, Psychology, Literature, Religion, Engineering, Mathematics, Food & Nutrition, Media & Journalism, Energy, Biotechnology, Sustainability, Agriculture, Human Rights, Urban Planning, Transportation, Immigration, Relationships & Family
+
+Exhibit Theme
+Natural History, Modern Art, Space Exploration, Ancient Egypt, Dinosaurs, Renaissance Art, Marine Life, Ancient Civilizations, Photography, Science and Technology, Contemporary Art, Medieval Europe, Roman Empire, Indigenous Cultures, World War II, Archaeology, Industrial Revolution, Fashion and Costume, Local History, Interactive Science, Human Evolution, Botany, Environmental Conservation, Maritime History, Textiles, Children's Exhibits, Music and Performing Arts, Sports History, Early Settlement, Design and Architecture, Medical History, Cultural Heritage
+
+Venue Type
+Bar, Restaurant, Club, Theater, Cinema, Concert Hall, Auditorium, Stadium, Arena, Convention/Conference Center, Amphitheater, Ballroom, Banquet Hall, Museum, Gallery, Casino, Sports Field, Sports Complex, Community Center, Park, Plaza, Rooftop Venue, Campus Venue, House of Worship, Private Residence, Comedy Club, Bowling Alley, Boat/Ship, Library, Racecourse
+
+Toy Type
+Construction & Building Sets, Dolls, Plush Toys (Stuffed Animals), Action Figures, Puzzles, Board Games, Toy Vehicles, Educational & STEM Toys, Pretend Play & Dress-up, Arts & Crafts Kits, Electronic & Interactive Toys, Remote Control Vehicles, Playsets (dollhouses, themed sets), Infant & Toddler Toys (stacking, shape sorters, lacing), Musical Toy Instruments, Ride-on Toys, Fidget & Sensory Toys, Trains & Train Sets, Marble Runs & Ball-Drop Toys, Model Kits & Hobby Sets, Die-cast Vehicles & Collectible Cars, Outdoor & Sports Toys, Bath Toys, Puppets, Card & Collectible Games
+
+Harvesting Method
+Mechanized harvesting, Hand harvesting, Combine harvester, Forage harvester, Baling, Windrowing, Tree shaker, Selective harvesting
+
+Hazard Mitigation Project Type
+Stormwater management, Levee rehabilitation and upgrade, Seawall construction and repair, Wetland restoration and creation, Floodplain restoration, Living shorelines, Beach nourishment, Reservoir construction, Storm surge barriers and gates, Pumping stations (stormwater/sea), Retention and detention basins, Dredging and sediment management, River channel restoration, Diversion channel construction, Culvert and bridge modifications, Bank stabilization, Coastal barrier enhancement, Breakwater construction, Groin construction, Dam removal (river restoration), Permeable pavement, Bioswales and rain gardens, Green roofs, Reforestation and afforestation, Slope stabilization (retaining walls, terraces), Floodproofing and structure elevation, Property acquisition and buyouts (managed retreat), Land-use planning and zoning measures, Early warning and monitoring systems, Evacuation route and emergency access improvements, Wildfire fuel reduction and prescribed burning, Firebreaks and defensible space
+
+Accessibility Feature Type
+Elevators, Ramps, Accessible parking spaces, Automatic doors, Wide doorways and corridors, Accessible restrooms, Curb ramps (curb cuts), Grab bars, Step-free / accessible routes, Braille signage, Tactile paving, Visual alarms (strobe alerts), Audio announcements, Hearing loop / induction loop systems, Lever door handles, Rocker light switches, Lowered countertops and service counters, Roll-in showers and accessible bathing, Vertical platform lifts and stairlifts, Accessible seating (priority and companion seating), High-contrast signage and markings, Accessible website / digital accessibility (WCAG), Captions and subtitles (video accessibility), Sign language interpretation services, Wayfinding and orientation features (including tactile maps)
+
+Tractor Powertrain & Control Configuration
+Diesel, Battery-electric, 2WD (two-wheel drive), 4WD (four-wheel drive), Tracked (crawler), Manual (operator-controlled), Auto-steer / GPS-assisted (semi-autonomous guidance), Autonomous (fully autonomous)
+
+Filtration Technology
+Activated Carbon Adsorption, Sand Filtration, Cartridge Filter, HEPA Filter, Reverse Osmosis, Microfiltration, Ultrafiltration, Nanofiltration, Ion Exchange, Filter Press
+
+Running Shoe Segment
+Neutral / Cushioned, Stability, Daily Trainer, Racing / Racing Flat, Carbon-Plated Racing Shoe, Trail Running, Minimalist / Barefoot, Maximalist / Max Cushion, Kids Running Shoe
+
+Gift Type
+Cash, Gift Card, Physical Gift, Digital Gift, Experience Gift, Subscription Gift, Flowers, Electronics, Clothing, Food & Beverage
+
+Farm Diversification Strategies
+Crop rotation, Crop diversification, Livestock integration, Direct-to-consumer sales, Value-added processing, Agritourism, Organic certification, Agroforestry, Renewable energy production, Community-supported agriculture (CSA)
+
+Payment Method
+Cash, Credit Card, Debit Card, Bank Transfer, Digital Wallet, PayPal, Check, Gift Card, Buy Now Pay Later, Cryptocurrency
+
+Climate Zone (major types — Köppen & common names)
+Tropical Rainforest, Tropical Savanna, Desert, Semi-arid (Steppe), Humid Subtropical, Oceanic (Marine West Coast), Mediterranean, Humid Continental, Tropical Monsoon, Temperate (general), Subarctic (Boreal), Tundra, Alpine (Highland), Polar (Arctic/Antarctic), Ice Cap
+
+Architectural Style
+Contemporary, Victorian, Craftsman, Colonial, Ranch, Mid-Century Modern, Farmhouse, Bungalow, Tudor, Mediterranean Revival, Spanish Colonial, Cape Cod, Art Deco, Industrial, Split-level, Neoclassical, Dutch Colonial, Log Cabin, Georgian, International Style, Federal, Mission Revival, Beaux-Arts, Prairie, Gothic Revival, Art Nouveau, Brutalist, Renaissance Revival
+
+Lighting Fixture Type
+Recessed Lighting, Ceiling Light, Flush Mount Ceiling Light, Pendant Light, Ceiling Fan, Chandelier, Wall Sconce, Table Lamp, Floor Lamp, Track Lighting, Vanity Light, Under-Cabinet Lighting, Linear Suspension, Semi-Flush Mount Ceiling Light, Spotlight, Floodlight, LED Strip Lighting, Accent Lighting, Picture Light, Landscape Lighting, Cove Lighting, Rope Light, Lantern (Outdoor Lantern), Path Light, Post Light, Solar Light, Bollard Light, Step Light, High Bay Light, Utility/Shop Light
+
+Washing Machine Type
+Front-load, Top-load agitator, Top-load impeller, Washer-dryer combo (all-in-one), Stackable washer-dryer (separate stacked units), Portable washing machine, Compact / undercounter washer, Semi-automatic twin-tub, Laundry center (single-unit stacked washer-dryer), Coin-operated / commercial washer, Industrial / high-capacity washer, Drawer washing machine
+
+Art Supply Set Type
+Colored Pencil Set, Watercolor Set, Acrylic Paint Set, Marker Set, Crayon Set, Graphite Pencil Set, Sketching Set, Charcoal Set, Oil Paint Set, Mixed Media Set
+
+Art & Craft Workshop Discipline
+Painting, Drawing, Photography, Sculpture, Ceramics, Textiles, Printmaking, Jewelry Making, Woodworking, Digital Art
+
+Scientific Discovery Category
+New Species, New Technology, New Medical Treatment, New Disease or Pathogen, Archaeological Site, Archaeological Artifact, New Fossil, Exoplanet, New Material, New Scientific Method or Technique
+
+Monetization Method
+Advertising, Subscriptions, Physical Product Sales, One-time Purchase, In-App Purchases, Freemium, Transaction Fees, Affiliate Marketing, Licensing and Royalties, Donations and Crowdfunding
+
+Debris Material
+Wood, Concrete, Metal, Plastic, Glass, Paper / Cardboard, Soil / Dirt, Asphalt, Electronic waste, Vegetation (green waste)
+
+Fishing Capture Method
+Trawl, Purse seine, Gillnet, Longline, Pots and traps, Trolling, Handline, Pole-and-line, Dredge, Spearfishing
+
+Land Management Practice
+Crop rotation, Cover cropping, Integrated pest management, Irrigation management, Conservation tillage, Reforestation, Agroforestry, Rotational grazing, Wetland restoration, Prescribed burning
+
+Basketball Shot Zone
+Restricted Area, In the Paint (Non-RA), Above the Break 3, Top of Key 3, Left Wing 3, Right Wing 3, Left Corner 3, Right Corner 3, Mid-Range, Elbow (Mid-Range), Long Two, Backcourt / Heave
+
+Dress Silhouette
+A-line, Sheath, Fit-and-flare, Wrap dress, Shift, Empire waist, Ball gown, Mermaid, Bodycon, Slip dress
+
+Water Conservation Measure
+Water-efficient fixtures, Leak detection and repair, Smart water meters, Drought-tolerant landscaping, Rainwater harvesting, Drip irrigation, Greywater reuse, Water audits, Rebate and incentive programs, Water use restrictions and watering schedules
+

icon_generation/image_gen.py

@@ -0,0 +1,119 @@
+import os
+import json
+from openai import OpenAI
+from PIL import Image
+from io import BytesIO
+import base64
+import sys
+import time
+import multiprocessing
+from concurrent.futures import ProcessPoolExecutor
+
+"The output must be a single image with an exact 6:4 aspect ratio (landscape orientation, width greater than height). The image must contain exactly 24 icons arranged in a strict grid of 6 columns (horizontal, left to right) and 4 rows (vertical, top to bottom). Do not rotate, transpose, or alter the grid orientation. No text, letters, numbers, labels, or captions. Use a pure white background only."
+
+API_KEY = os.getenv("OPENAI_API_KEY") or os.getenv("AIHUBMIX_API_KEY", "")
+API_PROVIDER = os.getenv("OPENAI_BASE_URL", "https://aihubmix.com/v1")
+
+# 创建OpenAI客户端的函数，每个进程需要自己的客户端实例
+def create_client():
+    return OpenAI(
+        api_key=API_KEY,
+        base_url=API_PROVIDER,
+    )
+
+# 全局客户端仅用于主进程
+client = create_client()
+
+def generate_image(task):
+    """生成指定attribute和value的图像"""
+    attribute, value = task
+    
+    # 每个进程创建自己的客户端实例
+    local_client = create_client()
+    
+    prompt = f"""Create a flat-design colorful pictogram symbolizing '{attribute}: {value}'. The design should be simple, no text or intricate details, no shading or gradients, and set against a white background."""
+    
+    # 检查图像是否已存在
+    output_path = os.path.join("images", f"{attribute}-{value}.png")
+    if os.path.exists(output_path):
+        print(f"图像已存在，跳过生成: {output_path}")
+        return None
+    
+    try:
+        result = local_client.images.generate(
+            model="gpt-image-1",
+            prompt=prompt,
+            n=1,
+            size="1024x1024",
+            quality="low",
+            moderation="low",
+            background="auto",
+        )
+        
+        print(f"生成 {attribute}: {value} 的图像成功")
+        
+        # 立即保存图像
+        if result and result.data:
+            image_base64 = result.data[0].b64_json
+            if image_base64:
+                image_bytes = base64.b64decode(image_base64)
+                with open(output_path, "wb") as f:
+                    f.write(image_bytes)
+                print(f"图片已保存至：{output_path}")
+                return True
+        
+        return None
+    except Exception as e:
+        print(f"生成 {attribute}: {value} 的图像失败: {e}")
+        return None
+
+def main():
+    # 读取filtered.json文件
+    try:
+        with open("filtered.json", "r", encoding="utf-8") as f:
+            data = json.load(f)
+    except Exception as e:
+        print(f"读取filtered.json失败: {e}")
+        return
+    
+    # 统计attribute-value对的总数
+    total_pairs = 0
+    attribute_counts = {}
+    for item in data:
+        attribute = item["name"]
+        value_count = len(item["values"])
+        total_pairs += value_count
+        attribute_counts[attribute] = value_count
+    
+    print(f"总共发现 {total_pairs} 个attribute-value对")
+    
+    # 确认是否继续
+    user_input = input("确认开始生成图像? (y/n): ")
+    if user_input.lower() != 'y':
+        print("已取消生成")
+        return
+    
+    # 创建输出目录
+    os.makedirs("images", exist_ok=True)
+    
+    # 创建任务列表
+    tasks = []
+    for item in data:
+        attribute = item["name"]
+        for value_info in item["values"]:
+            value = value_info["value"]
+            tasks.append((attribute, value))
+    
+    # 设置进程数量，根据CPU核心数量确定
+    num_processes = min(8, multiprocessing.cpu_count())
+    print(f"使用 {num_processes} 个进程并行处理")
+    
+    # 使用进程池并行处理
+    with ProcessPoolExecutor(max_workers=num_processes) as executor:
+        executor.map(generate_image, tasks)
+    
+    print("所有图像生成任务已完成")
+
+if __name__ == "__main__":
+    multiprocessing.freeze_support()  # Windows系统需要
+    main()

icon_generation/refine_domains.py

@@ -0,0 +1,537 @@
+#!/usr/bin/env python3
+"""
+使用 LLM 对 filtered.json 中的 name 字段进行 refine，
+将其改为更明确的 domain，并按 count 排序
+"""
+
+import json
+import os
+import requests
+from typing import Dict, Optional, List
+from collections import Counter
+from concurrent.futures import ThreadPoolExecutor, as_completed
+import threading
+
+
+class DomainRefiner:
+    """使用 LLM 来 refine domain names"""
+    
+    def __init__(self, api_key=None, base_url=None, model=None):
+        """
+        初始化 LLM analyzer
+        
+        Args:
+            api_key: API key
+            base_url: API base URL
+            model: Model name
+        """
+        self.api_key = api_key or os.getenv("OPENAI_API_KEY") or os.getenv("AIHUBMIX_API_KEY", "")
+        self.base_url = base_url or os.getenv("OPENAI_BASE_URL", "https://aihubmix.com/v1")
+        self.model = model or os.getenv("OPENAI_MODEL", "gemini-2.5-flash")
+        self.lock = threading.Lock()  # 线程锁，用于打印
+    
+    def filter_values(self, domain_name: str, values_with_counts: List[Dict]) -> List[Dict]:
+        """
+        使用 LLM 过滤掉不属于该 domain 的 specific values 和相似/重复的 values
+        
+        Args:
+            domain_name: domain 名称
+            values_with_counts: 包含 value 和 count 的列表 [{"value": "...", "count": ...}, ...]
+        
+        Returns:
+            过滤后的 values 列表
+        """
+        # 如果值太多，只发送 top 50 给 LLM
+        values_to_check = values_with_counts[:50] if len(values_with_counts) > 50 else values_with_counts
+        
+        prompt = self._build_filter_prompt(domain_name, values_to_check)
+        
+        try:
+            response = self._query_llm(prompt)
+            
+            if response:
+                # 清理可能的 markdown 代码块
+                cleaned_response = response.strip()
+                if cleaned_response.startswith('```'):
+                    lines = cleaned_response.split('\n')
+                    cleaned_response = '\n'.join(lines[1:-1] if lines[-1].strip() == '```' else lines[1:])
+                    cleaned_response = cleaned_response.replace('```json', '').replace('```', '').strip()
+                
+                result = json.loads(cleaned_response)
+                
+                # 验证返回格式
+                if 'filtered_values' in result:
+                    filtered_values_set = set(result['filtered_values'])
+                    
+                    # 对于 top 50，过滤它们
+                    filtered_top = [v for v in values_to_check if v['value'] in filtered_values_set]
+                    
+                    # 如果原始列表更长，保留剩余的（因为没有检查）
+                    if len(values_with_counts) > 50:
+                        remaining = values_with_counts[50:]
+                        filtered_list = filtered_top + remaining
+                    else:
+                        filtered_list = filtered_top
+                    
+                    return filtered_list
+                else:
+                    with self.lock:
+                        print(f"      ⚠️  LLM 过滤响应缺少字段，保留原始值")
+                    return values_with_counts
+                    
+            else:
+                with self.lock:
+                    print(f"      ⚠️  LLM 过滤失败，保留原始值")
+                return values_with_counts
+                
+        except json.JSONDecodeError as e:
+            with self.lock:
+                print(f"      ⚠️  LLM 过滤响应不是有效的 JSON，保留原始值")
+            return values_with_counts
+        except Exception as e:
+            with self.lock:
+                print(f"      ⚠️  值过滤错误: {e}，保留原始值")
+            return values_with_counts
+    
+    def _build_filter_prompt(self, domain_name: str, values_with_counts: List[Dict]) -> str:
+        """构建值过滤的 prompt"""
+        
+        values_str = '\n'.join([f'  - "{v["value"]}" (count: {v["count"]})' for v in values_with_counts])
+        
+        prompt = f"""Given a domain name and its associated values, please filter out:
+1. Values that don't truly belong to this domain (too specific, off-topic, or irrelevant)
+2. Similar or duplicate values (keep the most common or representative one)
+3. Values that are too generic or ambiguous
+
+Domain: "{domain_name}"
+
+Values to filter:
+{values_str}
+
+Please analyze these values and return ONLY the values that:
+- Clearly belong to this domain
+- Are distinct (not duplicates or very similar)
+- Are meaningful attributes
+
+Return your response in the following JSON format ONLY (no markdown, no extra text):
+{{
+    "filtered_values": ["value1", "value2", ...],
+    "removed_count": number_of_removed_values,
+    "reasoning": "Brief explanation of filtering criteria used"
+}}
+
+Examples:
+- Domain "Movie Genre" with values ["Action", "action movie", "ACT"] → Keep only "Action"
+- Domain "Country" with values ["USA", "New York", "California"] → Remove "New York", "California" (cities, not countries)
+"""
+        
+        return prompt
+    
+    def refine_domain_name(self, name: str, sample_values: List[str], total_count: int) -> Dict:
+        """
+        使用 LLM 分析并 refine domain name
+        
+        Args:
+            name: 原始 name
+            sample_values: 一些示例 values
+            total_count: 总计数
+        
+        Returns:
+            {
+                'original_name': str,
+                'refined_domain': str,
+                'reasoning': str
+            }
+        """
+        prompt = self._build_domain_refinement_prompt(name, sample_values, total_count)
+        
+        try:
+            response = self._query_llm(prompt)
+            
+            if response:
+                # 清理可能的 markdown 代码块
+                cleaned_response = response.strip()
+                if cleaned_response.startswith('```'):
+                    lines = cleaned_response.split('\n')
+                    cleaned_response = '\n'.join(lines[1:-1] if lines[-1].strip() == '```' else lines[1:])
+                    cleaned_response = cleaned_response.replace('```json', '').replace('```', '').strip()
+                
+                result = json.loads(cleaned_response)
+                
+                # 验证返回格式
+                if 'refined_domain' in result:
+                    result['original_name'] = name
+                    return result
+                else:
+                    with self.lock:
+                        print(f"      ❌ LLM 响应缺少必需字段: {result}")
+                    return None
+                    
+            else:
+                with self.lock:
+                    print("      ❌ LLM API 调用失败")
+                return None
+                
+        except json.JSONDecodeError as e:
+            with self.lock:
+                print(f"      ❌ LLM 响应不是有效的 JSON: {e}")
+                print(f"      响应内容: {response[:500]}...")
+            return None
+        except Exception as e:
+            with self.lock:
+                print(f"      ❌ Domain refinement 错误: {e}")
+            return None
+    
+    def _build_domain_refinement_prompt(self, name: str, sample_values: List[str], total_count: int) -> str:
+        """构建 domain refinement 的 prompt"""
+        
+        sample_values_str = ', '.join(f'"{v}"' for v in sample_values[:10])
+        
+        prompt = f"""Given a data field name and its sample values, please refine the name to a more specific and clear domain name.
+
+The domain name should:
+1. Clearly indicate what category/dimension this field represents
+2. Be consistent and professional
+3. Form a clear "domain: specific attribute" relationship with its values
+4. Be concise (1-3 words)
+
+Input Information:
+- Original Field Name: "{name}"
+- Sample Values: {sample_values_str}
+- Total Entries Count: {total_count}
+
+Please analyze the field name and sample values, then provide:
+1. A refined domain name that better describes this dimension
+2. Brief reasoning for your choice
+
+Return your response in the following JSON format ONLY (no markdown, no extra text):
+{{
+    "refined_domain": "YourRefinedDomainName",
+    "reasoning": "Brief explanation of why this domain name is more appropriate"
+}}
+
+Examples:
+- Original: "Genre" with values ["Action", "Rock", "Pop"] → Refined: "Entertainment Genre"
+- Original: "Type" with values ["Movie", "TV Show"] → Refined: "Media Type"
+- Original: "Category" with values ["Electronics", "Books"] → Refined: "Product Category"
+"""
+        
+        return prompt
+    
+    def _query_llm(self, prompt: str) -> Optional[str]:
+        """
+        查询 LLM API
+        
+        Args:
+            prompt: 发送给 LLM 的 prompt
+        
+        Returns:
+            str: LLM 响应内容
+        """
+        headers = {
+            'Authorization': f'Bearer {self.api_key}',
+            'Content-Type': 'application/json'
+        }
+        
+        data = {
+            'model': self.model,
+            'messages': [
+                {
+                    'role': 'system', 
+                    'content': 'You are a data modeling expert specialized in creating clear, semantic domain names. Always return valid JSON format only, without any markdown formatting or extra text.'
+                },
+                {
+                    'role': 'user', 
+                    'content': prompt
+                }
+            ],
+            'temperature': 0.3
+        }
+        
+        try:
+            response = requests.post(
+                f'{self.base_url}/chat/completions',
+                headers=headers,
+                json=data,
+                timeout=30
+            )
+            response.raise_for_status()
+            
+            result = response.json()
+            return result['choices'][0]['message']['content'].strip()
+            
+        except requests.exceptions.Timeout:
+            with self.lock:
+                print("❌ LLM API 超时")
+            return None
+        except requests.exceptions.HTTPError as e:
+            with self.lock:
+                print(f"❌ LLM API HTTP 错误: {e}")
+                if hasattr(e.response, 'text'):
+                    print(f"   响应: {e.response.text[:200]}")
+            return None
+        except requests.exceptions.RequestException as e:
+            with self.lock:
+                print(f"❌ LLM API 请求错误: {e}")
+            return None
+        except KeyError as e:
+            with self.lock:
+                print(f"❌ LLM API 响应格式错误: {e}")
+            return None
+
+
+def process_single_item(item: Dict, idx: int, total: int, refiner: DomainRefiner, start_time: float, start_idx: int) -> Dict:
+    """
+    处理单个数据项
+    
+    Args:
+        item: 数据项
+        idx: 当前索引
+        total: 总数量
+        refiner: DomainRefiner 实例
+        start_time: 开始时间
+        start_idx: 起始索引
+    
+    Returns:
+        处理后的数据项
+    """
+    import time
+    
+    # 计算进度信息
+    progress = (idx + 1) / total * 100
+    elapsed = time.time() - start_time
+    avg_time = elapsed / (idx - start_idx + 1) if idx > start_idx else 0
+    remaining = avg_time * (total - idx - 1)
+    
+    with refiner.lock:
+        print(f"\n{'=' * 80}")
+        print(f"🔄 处理进度: {idx + 1}/{total} ({progress:.1f}%)")
+        print(f"⏱️  已用时间: {elapsed:.1f}秒 | 预计剩余: {remaining:.1f}秒")
+        print(f"📝 当前字段: {item['name']}")
+        print(f"   - 值数量: {item['num_values']}")
+        print(f"   - 总计数: {item['total_count']}")
+    
+    # 提取 sample values
+    sample_values = [v['value'] for v in item['values'][:15]]
+    
+    with refiner.lock:
+        print(f"   - 示例值: {', '.join(sample_values[:5])}")
+        print(f"🤖 调用 LLM 进行 domain refinement...")
+    
+    # 使用 LLM refine domain name
+    refined_result = refiner.refine_domain_name(
+        item['name'],
+        sample_values,
+        item['total_count']
+    )
+    
+    if refined_result:
+        refined_domain = refined_result['refined_domain']
+        reasoning = refined_result.get('reasoning', '')
+        
+        with refiner.lock:
+            print(f"✅ Domain refinement 成功!")
+            print(f"   原始名称: '{item['name']}'")
+            print(f"   优化域名: '{refined_domain}'")
+            print(f"   优化理由: {reasoning}")
+            print(f"🔍 调用 LLM 进行 value filtering...")
+        
+        # 过滤 values
+        filtered_values = refiner.filter_values(refined_domain, item['values'])
+        
+        with refiner.lock:
+            removed_count = len(item['values']) - len(filtered_values)
+            print(f"✅ Value filtering 完成!")
+            print(f"   原始值数量: {len(item['values'])}")
+            print(f"   过滤后数量: {len(filtered_values)}")
+            print(f"   移除数量: {removed_count}")
+        
+        # 构建新的数据项
+        refined_item = {
+            'domain': refined_domain,
+            'original_name': item['name'],
+            'num_values': len(filtered_values),
+            'original_num_values': item['num_values'],
+            'total_count': sum(v['count'] for v in filtered_values),
+            'original_total_count': item['total_count'],
+            'values': filtered_values,
+            'refinement_reasoning': reasoning
+        }
+        
+        return refined_item
+    else:
+        # 如果 LLM 失败，保留原始 name 作为 domain
+        with refiner.lock:
+            print(f"⚠️  LLM refinement 失败，使用原始 name，跳过值过滤")
+        
+        refined_item = {
+            'domain': item['name'],
+            'original_name': item['name'],
+            'num_values': item['num_values'],
+            'original_num_values': item['num_values'],
+            'total_count': item['total_count'],
+            'original_total_count': item['total_count'],
+            'values': item['values'],
+            'refinement_reasoning': 'LLM refinement failed, kept original'
+        }
+        
+        return refined_item
+
+
+def process_filtered_json(input_file: str, output_file: str, temp_file: str = None, num_threads: int = 10):
+    """
+    处理 filtered.json 文件（并行处理）
+    
+    Args:
+        input_file: 输入文件路径
+        output_file: 输出文件路径
+        temp_file: 临时文件路径，用于实时保存中间结果
+        num_threads: 并行线程数
+    """
+    import os
+    import time
+    
+    if temp_file is None:
+        temp_file = output_file.replace('.json', '_temp.json')
+    
+    print(f"📖 读取文件: {input_file}")
+    print(f"💾 临时文件: {temp_file}")
+    print(f"✨ 最终文件: {output_file}")
+    print(f"🔧 并行线程数: {num_threads}")
+    
+    # 读取原始数据
+    with open(input_file, 'r', encoding='utf-8') as f:
+        data = json.load(f)
+    
+    print(f"✅ 成功读取 {len(data)} 个字段\n")
+    print("=" * 80)
+    
+    # 检查是否已有临时文件（支持断点续传）
+    refined_data = []
+    start_idx = 0
+    processed_names = set()
+    
+    if os.path.exists(temp_file):
+        print(f"📂 发现临时文件，尝试恢复进度...")
+        try:
+            with open(temp_file, 'r', encoding='utf-8') as f:
+                refined_data = json.load(f)
+            start_idx = len(refined_data)
+            processed_names = {item['original_name'] for item in refined_data}
+            print(f"✅ 已恢复 {start_idx} 个字段的处理结果")
+        except Exception as e:
+            print(f"⚠️  临时文件读取失败: {e}，从头开始")
+            refined_data = []
+            start_idx = 0
+            processed_names = set()
+    
+    # 过滤掉已处理的项
+    items_to_process = [item for item in data if item['name'] not in processed_names]
+    
+    if not items_to_process:
+        print("✅ 所有项目已处理完成！")
+        return
+    
+    print(f"📋 待处理项目: {len(items_to_process)} 个")
+    print("=" * 80)
+    
+    # 初始化 LLM refiner
+    refiner = DomainRefiner()
+    
+    # 记录开始时间
+    start_time = time.time()
+    
+    # 使用线程池并行处理
+    with ThreadPoolExecutor(max_workers=num_threads) as executor:
+        # 提交所有任务
+        future_to_idx = {
+            executor.submit(
+                process_single_item, 
+                item, 
+                start_idx + i, 
+                len(data), 
+                refiner, 
+                start_time, 
+                start_idx
+            ): (start_idx + i, item)
+            for i, item in enumerate(items_to_process)
+        }
+        
+        # 按完成顺序收集结果
+        completed = 0
+        for future in as_completed(future_to_idx):
+            idx, item = future_to_idx[future]
+            try:
+                result = future.result()
+                refined_data.append(result)
+                completed += 1
+                
+                # 每处理 5 个项目保存一次
+                if completed % 5 == 0:
+                    with refiner.lock:
+                        print(f"\n{'=' * 80}")
+                        print(f"💾 保存中间结果... (已完成 {completed}/{len(items_to_process)})")
+                    with open(temp_file, 'w', encoding='utf-8') as f:
+                        json.dump(refined_data, f, ensure_ascii=False, indent=2)
+                    with refiner.lock:
+                        print(f"✅ 已保存 {len(refined_data)} 个字段到临时文件")
+                
+            except Exception as e:
+                with refiner.lock:
+                    print(f"\n❌ 处理项目 {idx} ({item['name']}) 时出错: {e}")
+    
+    # 最终保存一次
+    print(f"\n{'=' * 80}")
+    print(f"💾 保存最终临时结果...")
+    with open(temp_file, 'w', encoding='utf-8') as f:
+        json.dump(refined_data, f, ensure_ascii=False, indent=2)
+    
+    # 按 total_count 降序排序
+    print(f"\n{'=' * 80}")
+    print("📊 按 total_count 进行排序...")
+    refined_data.sort(key=lambda x: x['total_count'], reverse=True)
+    
+    # 保存最终结果
+    print(f"💾 保存最终结果到: {output_file}")
+    with open(output_file, 'w', encoding='utf-8') as f:
+        json.dump(refined_data, f, ensure_ascii=False, indent=2)
+    
+    print(f"\n{'=' * 80}")
+    print(f"✅ 处理完成！共处理 {len(refined_data)} 个字段")
+    print(f"⏱️  总耗时: {time.time() - start_time:.1f}秒")
+    
+    # 输出统计信息
+    print(f"\n{'=' * 80}")
+    print("📈 统计信息:")
+    print(f"   总字段数: {len(refined_data)}")
+    print(f"   总记录数（过滤后）: {sum(item['total_count'] for item in refined_data):,}")
+    print(f"   总记录数（原始）: {sum(item.get('original_total_count', item['total_count']) for item in refined_data):,}")
+    
+    # 计算过滤统计
+    total_values_before = sum(item.get('original_num_values', item['num_values']) for item in refined_data)
+    total_values_after = sum(item['num_values'] for item in refined_data)
+    print(f"   总值数量（原始）: {total_values_before:,}")
+    print(f"   总值数量（过滤后）: {total_values_after:,}")
+    print(f"   过滤比例: {(1 - total_values_after/total_values_before)*100:.1f}%")
+    
+    # 显示前 10 个 domain
+    print(f"\n{'=' * 80}")
+    print("🏆 Top 10 Domains (按 total_count):")
+    for i, item in enumerate(refined_data[:10]):
+        print(f"\n   {i+1}. {item['domain']} (原: {item['original_name']})")
+        print(f"      Count: {item['total_count']:,}, Values: {item['num_values']}")
+        print(f"      示例值: {', '.join([v['value'] for v in item['values'][:5]])}")
+    
+    # 删除临时文件
+    if os.path.exists(temp_file):
+        print(f"\n🗑��  保留临时文件以备恢复: {temp_file}")
+        # os.remove(temp_file)  # 暂时不删除，以便需要时恢复
+
+
+if __name__ == '__main__':
+    input_file = '/home/lizhen/ChartPipeline/icon_generation/filtered.json'
+    output_file = '/home/lizhen/ChartPipeline/icon_generation/refined_domains.json'
+    
+    process_filtered_json(input_file, output_file, num_threads=10)
+
+

icon_generation/split_icon.py

@@ -0,0 +1,487 @@
+import cv2
+import numpy as np
+import os
+import re
+import glob
+
+def clean_filename(text):
+    """清理文件名"""
+    text = text.lower().strip()
+    text = re.sub(r'[\s\W]+', '_', text)
+    return text.strip('_')
+
+def sort_contours(cnts, method="left-to-right"):
+    """
+    对轮廓进行排序。
+    对于 Grid 布局，我们需要 'top-to-bottom' 然后 'left-to-right' 的混合排序。
+    """
+    if not cnts:
+        return [], []
+
+    # 获取每个轮廓的 Bounding Box
+    boundingBoxes = [cv2.boundingRect(c) for c in cnts]
+    
+    # 将轮廓和bbox打包
+    cnts_boxes = list(zip(cnts, boundingBoxes))
+    
+    # 1. 按照 Y 坐标（从上到下）进行初步排序
+    # key: y
+    cnts_boxes.sort(key=lambda b: b[1][1])
+    
+    # 2. 分行处理
+    # 由于手工画线或扫描误差，同一行的y坐标可能不完全相同。
+    # 我们需要设定一个阈值，认为y坐标相近的是“同一行”。
+    rows = []
+    current_row = []
+    if cnts_boxes:
+        # 以第一个轮廓的高度作为参考阈值
+        ref_h = cnts_boxes[0][1][3]
+        tolerance = ref_h * 0.5 # 容差设为高度的一半
+        
+        last_y = cnts_boxes[0][1][1]
+        
+        for c, box in cnts_boxes:
+            y = box[1]
+            if y <= last_y + tolerance:
+                current_row.append((c, box))
+            else:
+                # 新的一行
+                rows.append(current_row)
+                current_row = [(c, box)]
+                last_y = y
+        # 添加最后一行
+        if current_row:
+            rows.append(current_row)
+
+    # 3. 对每一行内部，按照 X 坐标（从左到右）排序
+    final_sorted = []
+    row_counts = []
+    for i, row in enumerate(rows):
+        # key: x
+        row.sort(key=lambda b: b[1][0])
+        row_counts.append(len(row))
+        for item in row:
+            final_sorted.append(item[1]) # 只返回 bbox (x, y, w, h)
+
+    return final_sorted, row_counts
+
+def uniform_grid_split(img, expected_cols=6, expected_rows=4, margin_percent=0.02):
+    """
+    均匀分割方法：直接按照预期的行列数均匀分割图像
+    适用于网格线不连续或没有明显网格线的情况
+    
+    Args:
+        img: 输入图像
+        expected_cols: 期望的列数
+        expected_rows: 期望的行数
+        margin_percent: 边缘裁剪比例（去除可能的边框）
+    
+    Returns:
+        排序好的 (x, y, w, h) 列表
+    """
+    h_img, w_img = img.shape[:2]
+    
+    # 去除边缘
+    margin_x = int(w_img * margin_percent)
+    margin_y = int(h_img * margin_percent)
+    
+    effective_width = w_img - 2 * margin_x
+    effective_height = h_img - 2 * margin_y
+    
+    # 计算每个单元格的尺寸
+    cell_width = effective_width // expected_cols
+    cell_height = effective_height // expected_rows
+    
+    boxes = []
+    for row in range(expected_rows):
+        for col in range(expected_cols):
+            x = margin_x + col * cell_width
+            y = margin_y + row * cell_height
+            boxes.append((x, y, cell_width, cell_height))
+    
+    return boxes
+
+def detect_grid_cells_with_lines(img, expected_cols=6, expected_rows=4):
+    """
+    通过形态学操作检测网格线，并提取每个格子的坐标
+    """
+    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+    
+    # 二值化 (反转：背景黑，内容/线白)
+    # 使用自适应阈值来应对光照或颜色不均
+    thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
+                                   cv2.THRESH_BINARY_INV, 11, 2)
+
+    # 定义结构元素 (Kernel) - 增大kernel以更好地检测断裂的线
+    h_img, w_img = img.shape[:2]
+    # 水平线 Kernel: 宽度长，高度为1
+    horizontal_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (w_img // 15, 1))
+    # 垂直线 Kernel: 宽度为1，高度长
+    vertical_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, h_img // 15))
+
+    # 1. 提取水平线
+    detect_horizontal = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, horizontal_kernel, iterations=2)
+    
+    # 2. 提取垂直线
+    detect_vertical = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, vertical_kernel, iterations=2)
+
+    # 3. 合并网格线
+    grid_mask = cv2.addWeighted(detect_horizontal, 0.5, detect_vertical, 0.5, 0)
+    _, grid_mask = cv2.threshold(grid_mask, 0, 255, cv2.THRESH_BINARY)
+    
+    # 更强的膨胀操作，连接断裂的网格线
+    kernel_dilate = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
+    grid_mask = cv2.dilate(grid_mask, kernel_dilate, iterations=3)
+    
+    # 闭运算，进一步连接断裂
+    kernel_close = cv2.getStructuringElement(cv2.MORPH_RECT, (7, 7))
+    grid_mask = cv2.morphologyEx(grid_mask, cv2.MORPH_CLOSE, kernel_close, iterations=2)
+
+    # 4. 寻找所有的“洞”（即单元格）
+    # 我们通过查找 grid_mask 的轮廓，通常很难直接找到内部的矩形。
+    # 更好的方法是：找出网格��轮廓，画在全黑背景上，然后寻找连通组件，或者反转图片找白色方块。
+    
+    # 这里我们采用“反转 mask”法：网格线是黑，格子是白
+    contours_mask = cv2.bitwise_not(grid_mask)
+    contours, _ = cv2.findContours(contours_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+
+    # 计算期望的单元格面积
+    expected_cell_area = (w_img * h_img) / (expected_cols * expected_rows)
+    
+    # 过滤微小的噪点轮廓，同时也过滤过大的轮廓
+    # 放宽过滤条件以捕获更多单元格
+    min_area = expected_cell_area / 20  # 单元格面积的 1/20 (之前是1/10)
+    max_area = expected_cell_area * 3   # 单元格面积的3倍 (之前是2倍)
+    
+    # 调试信息
+    print(f"    [调试] 图像尺寸: {w_img}x{h_img}, 检测到轮廓: {len(contours)}个")
+    print(f"    [调试] 期望单元格面积: {expected_cell_area:.0f}, 过滤范围: {min_area:.0f}-{max_area:.0f}")
+    
+    # 统计被过滤掉的轮廓
+    filtered_out = []
+    valid_contours = []
+    for c in contours:
+        area = cv2.contourArea(c)
+        if min_area < area < max_area:
+            valid_contours.append(c)
+        else:
+            x, y, w, h = cv2.boundingRect(c)
+            filtered_out.append((area, x, y, w, h))
+    
+    if filtered_out:
+        print(f"    [调试] 被过滤掉 {len(filtered_out)} 个轮廓:")
+        for area, x, y, w, h in sorted(filtered_out, key=lambda t: t[0], reverse=True)[:5]:
+            print(f"      - 面积={area:.0f}, 位置=({x},{y}), 尺寸={w}x{h}")
+
+    # 排序：确保顺序是 左->右，上->下
+    sorted_boxes, row_counts = sort_contours(valid_contours)
+    
+    return sorted_boxes, row_counts
+
+def detect_grid_cells(img, expected_cols=6, expected_rows=4):
+    """
+    鲁棒的网格检测方法：首先尝试检测网格线，如果失败则使用均匀分割
+    """
+    expected_count = expected_cols * expected_rows
+    
+    # 方法1: 尝试检测网格线
+    sorted_boxes, row_counts = detect_grid_cells_with_lines(img, expected_cols, expected_rows)
+    
+    # 严格检查：必须恰好检测到期望数量的单元格
+    if len(sorted_boxes) != expected_count:
+        print(f"  网格线检测不理想（检测到 {len(sorted_boxes)} 个单元格，期望 {expected_count} 个）")
+        if row_counts:
+            row_info = ", ".join([f"第{i+1}行: {count}个" for i, count in enumerate(row_counts)])
+            print(f"  检测到的行分布: {row_info}")
+        print(f"  切换到均匀分割模式...")
+        sorted_boxes = uniform_grid_split(img, expected_cols, expected_rows)
+        # 均匀分割时，打印每行的单元格数
+        print(f"  均匀分割结果：每行 {expected_cols} 个单元格，共 {expected_rows} 行")
+    else:
+        print(f"  ✓ 成功检测到 {len(sorted_boxes)} 个网格单元格（符合预期）")
+        # 打印每行的单元格数量
+        if row_counts:
+            row_info = ", ".join([f"第{i+1}行: {count}个" for i, count in enumerate(row_counts)])
+            print(f"  行分布: {row_info}")
+    
+    return sorted_boxes
+
+def is_likely_text_region(img_region, thresh_region):
+    """
+    判断一个区域是否可能是文字
+    文字的特征：
+    1. 主要是黑色或深色
+    2. 高度较小
+    3. 像素密度适中（不是纯色块）
+    """
+    if img_region.shape[0] == 0 or img_region.shape[1] == 0:
+        return False
+    
+    # 转换为灰度（如果不是）
+    if len(img_region.shape) == 3:
+        gray_region = cv2.cvtColor(img_region, cv2.COLOR_BGR2GRAY)
+    else:
+        gray_region = img_region
+    
+    # 检查1：高度不能太大（文字通常较矮）
+    height_ratio = img_region.shape[0] / img_region.shape[1] if img_region.shape[1] > 0 else 1
+    if height_ratio > 0.3:  # 如果高度超过宽度的30%，可能不是单行文字
+        return False
+    
+    # 检查2：颜色是否偏暗（文字通常是黑色或深色）
+    mean_brightness = np.mean(gray_region)
+    if mean_brightness > 200:  # 太亮，不像文字
+        return False
+    
+    # 检查3：内容像素占比（文字不会太密集也不会太稀疏）
+    content_pixels = np.sum(thresh_region > 0)
+    total_pixels = thresh_region.shape[0] * thresh_region.shape[1]
+    density = content_pixels / total_pixels if total_pixels > 0 else 0
+    
+    if density < 0.05 or density > 0.5:  # 密度不在合理范围
+        return False
+    
+    return True
+
+def detect_and_remove_text(img, thresh, row_sums):
+    """
+    检测并移除图标上方或下方的文字标题
+    
+    返回: (top_crop, bottom_crop) - 需要裁剪的上下边界
+    """
+    h = len(row_sums)
+    
+    # 定义"空白行"的阈值（行和很小）
+    empty_threshold = max(5, img.shape[1] * 0.01)  # 至少5，或宽度的1%
+    # 定义"间隙"的最小行数
+    min_gap_rows = max(2, int(h * 0.02))  # 至少2行，或高度的2%
+    
+    # 找到所有内容行（非空白行）
+    content_rows = [i for i, val in enumerate(row_sums) if val > empty_threshold]
+    
+    if len(content_rows) == 0:
+        return 0, h
+    
+    # 找到主要内容区域（最大的连续内容块）
+    # 先找出所有的间隙
+    gaps = []
+    if len(content_rows) > 1:
+        for i in range(len(content_rows) - 1):
+            gap_size = content_rows[i + 1] - content_rows[i] - 1
+            if gap_size >= min_gap_rows:
+                gap_start = content_rows[i]
+                gap_end = content_rows[i + 1]
+                gaps.append((gap_start, gap_end, gap_size))
+    
+    top_crop = 0
+    bottom_crop = h
+    
+    # 如果存在明显的间隙，说明可能有分离的文字
+    if gaps:
+        # 找到最大的间隙
+        largest_gap = max(gaps, key=lambda x: x[2])
+        gap_start, gap_end, gap_size = largest_gap
+        
+        # 计算间隙上方和下方的内容量和行数
+        top_rows = gap_start
+        bottom_rows = h - gap_end
+        top_content = sum(row_sums[:gap_start])
+        bottom_content = sum(row_sums[gap_end:])
+        
+        # 判断哪一部分是主要图标，哪一部分是文字
+        # 文字的特征：1) 内容较少 2) 行数较少 3) 符合文字特征
+        
+        # 检查上方区域
+        if top_rows > 0 and top_rows < h * 0.3:  # 上方行数不超过30%
+            if top_content < bottom_content * 0.4:  # 上方内容明显少于下方
+                # 进一步检查是否像文字
+                top_region = img[:gap_start, :]
+                top_thresh = thresh[:gap_start, :]
+                if is_likely_text_region(top_region, top_thresh):
+                    top_crop = gap_end
+        
+        # 检查下方区域
+        if bottom_rows > 0 and bottom_rows < h * 0.3:  # 下方行数不超过30%
+            if bottom_content < top_content * 0.4:  # 下方内容明显少于上方
+                # 进一步检查是否像文字
+                bottom_region = img[gap_end:, :]
+                bottom_thresh = thresh[gap_end:, :]
+                if is_likely_text_region(bottom_region, bottom_thresh):
+                    bottom_crop = gap_start
+    
+    return top_crop, bottom_crop
+
+def smart_crop_icon(img, padding=10):
+    """
+    单个 Icon 处理：去字、去空、加 Padding
+    增强版：可以检测并删除上方或下方的文字标题
+    """
+    h, w = img.shape[:2]
+    
+    # 1. 裁剪掉可能残留的网格边缘 (比如四周切掉 3px)
+    margin = 3
+    if h > 2*margin and w > 2*margin:
+        img = img[margin:-margin, margin:-margin]
+        h, w = img.shape[:2]
+    
+    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+    _, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
+
+    # 2. 使用改进的方法检测并移除文字
+    row_sums = np.sum(thresh, axis=1)
+    top_crop, bottom_crop = detect_and_remove_text(img, thresh, row_sums)
+    
+    # 应用裁剪
+    if top_crop > 0 or bottom_crop < h:
+        img = img[top_crop:bottom_crop, :]
+        thresh = thresh[top_crop:bottom_crop, :]
+
+    # 3. 寻找 Icon 的精确边界
+    coords = cv2.findNonZero(thresh)
+    if coords is not None:
+        x, y, w_box, h_box = cv2.boundingRect(coords)
+        
+        # 裁剪并添加 Padding
+        # 创建一个新的白色画布
+        final_h = h_box + 2 * padding
+        final_w = w_box + 2 * padding
+        canvas = np.ones((final_h, final_w, 3), dtype=np.uint8) * 255
+        
+        # 提取 icon 内容
+        icon_content = img[y:y+h_box, x:x+w_box]
+        
+        # 将 icon 贴到画布中心
+        canvas[padding:padding+h_box, padding:padding+w_box] = icon_content
+        return canvas
+    
+    return img
+
+def process_image_robust(image_path, labels_data):
+    if not os.path.exists(image_path):
+        print(f"Error: {image_path} not found.")
+        return
+
+    print(f"Processing: {image_path} ...")
+    img = cv2.imread(image_path)
+    
+    # 1. 检测网格
+    # 返回的是排序好的 (x, y, w, h) 列表
+    grid_boxes = detect_grid_cells(img, expected_cols=6, expected_rows=4)
+    
+    # 2. 准备文本数据
+    lines = [l.strip() for l in labels_data.strip().split('\n') if l.strip()]
+    style = "flat"
+    start_idx = 0
+    if lines[0].lower().startswith("style:"):
+        style = clean_filename(lines[0].split(':')[1])
+        start_idx = 1
+        
+    output_dir = "extracted_icons"
+    if not os.path.exists(output_dir):
+        os.makedirs(output_dir)
+
+    # 3. 遍历并保存
+    for i, box in enumerate(grid_boxes):
+        text_idx = start_idx + i
+        if text_idx >= len(lines):
+            break
+            
+        # 解析文本
+        line_text = lines[text_idx]
+        parts = line_text.split(',', 1)
+        if len(parts) == 2:
+            category = clean_filename(parts[0])
+            name = clean_filename(parts[1])
+        else:
+            category = "icon"
+            name = clean_filename(parts[0])
+            
+        filename = f"{category}-{name}-{style}.png"
+        save_path = os.path.join(output_dir, filename)
+        
+        # 提取单元格
+        x, y, w, h = box
+        cell_img = img[y:y+h, x:x+w]
+        
+        # 智能裁切
+        final_img = smart_crop_icon(cell_img, padding=10)
+        
+        cv2.imwrite(save_path, final_img)
+        # print(f"Saved: {filename}") # 减少刷屏
+
+    print(f"Done. Extracted {len(grid_boxes)} icons to '{output_dir}/'.\n")
+
+def process_batch_range(start_batch, end_batch, base_dir="generated_icons"):
+    """
+    批量处理指定范围内的batch文件夹下的所有png文件
+    
+    Args:
+        start_batch: 起始batch编号 (例如: 1)
+        end_batch: 结束batch编号 (例如: 10)
+        base_dir: batch文件夹所在的基础目录
+    """
+    print(f"开始批量处理 batch_{start_batch:04d} 到 batch_{end_batch:04d} ...\n")
+    
+    total_processed = 0
+    failed_files = []
+    
+    for batch_num in range(start_batch, end_batch + 1):
+        batch_dir = os.path.join(base_dir, f"batch_{batch_num:04d}")
+        
+        # 检查batch文件夹是否存在
+        if not os.path.exists(batch_dir):
+            print(f"Warning: {batch_dir} 不存在，跳过...")
+            continue
+        
+        print(f"处理 {batch_dir} ...")
+        
+        # 查找该batch下的所有png文件
+        png_files = glob.glob(os.path.join(batch_dir, "*.png"))
+        
+        if not png_files:
+            print(f"  未找到png文件，跳过...")
+            continue
+        
+        # 处理每个png文件
+        for png_path in sorted(png_files):
+            # 构造对应的txt文件路径
+            txt_path = png_path.rsplit('.', 1)[0] + '.txt'
+            
+            # 检查txt文件是否存在
+            if not os.path.exists(txt_path):
+                print(f"  Warning: {txt_path} 不存在，跳过 {os.path.basename(png_path)}")
+                failed_files.append(png_path)
+                continue
+            
+            # 读取txt文件内容
+            try:
+                with open(txt_path, 'r', encoding='utf-8') as f:
+                    labels_data = f.read()
+                
+                # 处理图像
+                process_image_robust(png_path, labels_data)
+                total_processed += 1
+                
+            except Exception as e:
+                print(f"  Error processing {os.path.basename(png_path)}: {str(e)}")
+                failed_files.append(png_path)
+    
+    # 输出总结
+    print(f"\n{'='*60}")
+    print(f"批量处理完成！")
+    print(f"成功处理: {total_processed} 个文件")
+    
+    if failed_files:
+        print(f"失败/跳过: {len(failed_files)} 个文件")
+        print("失败文件列表:")
+        for f in failed_files:
+            print(f"  - {f}")
+    print(f"{'='*60}")
+
+if __name__ == "__main__":
+    # 设置要处理的batch范围
+    START_BATCH = 1    # 起始batch编号
+    END_BATCH = 200     # 结束batch编号
+    
+    # 执行批量处理
+    process_batch_range(START_BATCH, END_BATCH)

icon_generation/template.json

@@ -0,0 +1,14 @@
+{
+  "minimal_flat": "Create a minimal flat icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use simple flat shapes with solid colors, no outlines, no gradients, no shadows, no depth, no texture. Maintain a clean, modern look with clear geometry and a white background.",
+  "outline_line": "Create an outline icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use clean, uniform stroke lines only, no fills, no shading, no texture, no gradients. Keep the design light, airy, and minimal, with a white background.",
+  "solid_filled": "Create a solid filled icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use bold, fully filled shapes with strong silhouettes, no outlines, no gradients, no texture, no shading. High contrast, simple form, white background.",
+  "simplified_cartoon": "Create a simplified cartoon-style icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use rounded shapes, playful proportions, and simplified details. Avoid realism, texture, or shading. Keep the icon friendly, colorful, and clean on a white background.",
+  "hand_drawn_sketch": "Create a hand-drawn sketch-style icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use rough, imperfect strokes with visible hand-drawn variation. No straight mechanical lines, no fills, minimal shading, organic and expressive lines on a white background.",
+  "doodle": "Create a doodle-style icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use loose, casual, spontaneous linework with playful and whimsical energy. Keep it simple and informal, no precise geometry, no fills, no shading, white background.",
+  "isometric": "Create an isometric icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use a fixed isometric perspective with simple 3D-like forms. Apply flat colors with subtle separation between surfaces, no realistic lighting, no heavy shadows, clean and structured on a white background.",
+  "pictogram": "Create a pictogram icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use highly simplified, universally recognizable shapes with clear symbolism. Avoid decorative details, textures, or perspective. Flat, functional design on a white background.",
+  "glyph": "Create a glyph icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use extremely minimal, symbol-like forms designed for clarity at small sizes. Solid shapes only, no details, no texture, no shading, monochrome on a white background.",
+  "pixel": "Create a pixel-style icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use a visible pixel grid, low resolution, and blocky shapes. Avoid smooth curves, gradients, or anti-aliasing. Retro digital style on a plain background.",
+  "chalkboard": "Create a chalkboard-style icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use chalk-like textured strokes with slightly uneven edges. White or light chalk lines on a dark chalkboard background. No smooth vector lines, educational and hand-drawn feel.",
+  "neon_glow": "Create a neon glow icon representing {DOMAIN}: {SPECIFIC ATTRIBUTE}. Use bright glowing outlines with a soft neon light effect. Dark background, high contrast, futuristic or nightlife aesthetic. Avoid flat colors or solid fills."
+}

icon_generation/template_batch.json

@@ -0,0 +1,21 @@
+{
+  "minimal_flat": "Create a grid of minimal flat icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Use a clear and strict grid layout with thin black divider lines separating each icon. Each icon should have equal size and consistent alignment. Use simple flat shapes with solid colors, no outlines, no gradients, no shadows, no depth, no texture. Clean, modern geometry.",
+
+  "outline_line": "Create a grid of colored outline icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Use a strict grid layout with thin black divider lines between icons. All icons must be uniform in size and spacing. Use clean, consistent colored stroke lines only. No fills, no shading, no texture, no gradients. Icons should feature simple outlines rendered in color, not black.",
+
+  "solid_filled": "Create a grid of solid filled icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Arrange icons in a uniform grid separated by thin black lines. Each icon should have a strong silhouette, fully filled shapes, no outlines, no gradients, no texture, no shading. High contrast, simple forms.",
+
+  "simplified_cartoon": "Create a grid of simplified cartoon-style icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Use a consistent grid layout with thin black divider lines. Icons must be equal in size and evenly spaced. Use rounded shapes, playful proportions, simplified details. No realism, no texture, no shading.",
+
+  "hand_drawn_sketch": "Create a grid of hand-drawn sketch-style icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Display icons in a structured grid separated by thin black lines. Maintain consistent icon size despite sketch variation. Use rough, imperfect strokes with visible hand-drawn character. Apply appropriate colorful fills to each icon, while keeping minimal shading and organic lines.",
+
+  "doodle": "Create a grid of doodle-style icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Arrange icons in a clear grid with thin black divider lines. Keep all icons evenly sized and aligned. Use loose, playful, spontaneous linework with a casual feel, and add colorful fills to each icon. No shading, no precise geometry.",
+
+  "isometric": "Create a grid of isometric icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Use a strict grid layout with thin black divider lines between icons. Maintain consistent scale and isometric angle across all icons. Use flat colors with simple surface separation, no realistic lighting, no heavy shadows.",
+
+  "glyph": "Create a grid of glyph icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Use a precise grid layout with thin black divider lines. All glyphs must be equal in size and visually balanced. Use extremely minimal, symbol-like solid forms with minimal coloring—use just a few appropriate colors sparingly for clarity or emphasis. No details, no texture, no shading. Avoid monochrome.",
+
+  "chalkboard": "Create a grid of colorful chalk-style icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Use a fixed grid layout separated by thin black divider lines. All icons should be consistently sized. Use multicolored chalk-like textured strokes with slightly uneven edges, vibrant pastel colors, and a visible hand-drawn look. Adapt for high visibility and contrast on a white background.",
+
+  "neon_glow": "Create a grid of neon-style glow icons. Each icon represents a specific domain-attribute pair from the following list: {DOMAIN_ATTRIBUTE_PAIRS}. Arrange icons in a strict grid with thin black divider lines. Icons should be uniform in size and alignment. Use bright glowing outlines and soft neon effects adapted for high contrast on a white background."
+}

icon_generation/test_split.py

@@ -0,0 +1,56 @@
+#!/usr/bin/env python3
+"""
+测试split_icon.py的网格检测功能
+"""
+import cv2
+from split_icon import detect_grid_cells, uniform_grid_split
+import os
+
+def test_image(image_path):
+    """测试单个图像的网格检测"""
+    if not os.path.exists(image_path):
+        print(f"错误：文件不存在 {image_path}")
+        return
+    
+    print(f"\n{'='*60}")
+    print(f"测试图像: {os.path.basename(image_path)}")
+    print(f"{'='*60}")
+    
+    # 读取图像
+    img = cv2.imread(image_path)
+    if img is None:
+        print("错误：无法读取图像")
+        return
+    
+    print(f"图像尺寸: {img.shape[1]}x{img.shape[0]}")
+    
+    # 测试网格检测
+    boxes = detect_grid_cells(img, expected_cols=6, expected_rows=4)
+    print(f"最终检测到的单元格数量: {len(boxes)}")
+    
+    # 可视化结果
+    result_img = img.copy()
+    for i, box in enumerate(boxes):
+        x, y, w, h = box
+        cv2.rectangle(result_img, (x, y), (x+w, y+h), (0, 255, 0), 2)
+        # 在左上角添加序号
+        cv2.putText(result_img, str(i+1), (x+5, y+20), 
+                   cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 2)
+    
+    # 保存结果
+    output_path = image_path.rsplit('.', 1)[0] + '_detected.png'
+    cv2.imwrite(output_path, result_img)
+    print(f"检测结果已保存到: {output_path}")
+
+if __name__ == "__main__":
+    # 测试batch_0001中的一张图像
+    test_images = [
+        "generated_icons/batch_0001/batch_0001_doodle.png",
+        "generated_icons/batch_0001/batch_0001_hand_drawn_sketch.png",
+        "generated_icons/batch_0001/batch_0001_isometric.png",
+    ]
+    
+    for img_path in test_images:
+        if os.path.exists(img_path):
+            test_image(img_path)
+            break