Upload 8 files

DEPLOY_INSTRUCTIONS.md

@@ -0,0 +1,71 @@
+# Hugging Face Spaces 部署说明
+
+## 📁 需要上传的文件
+
+这个文件夹包含所有需要推送到 Hugging Face Spaces 的文件：
+
+```
+hf_deploy/
+├── app.py                  # FastAPI 应用（已更新 max_tokens: 4000）
+├── Dockerfile              # Docker 配置
+├── requirements.txt        # Python 依赖
+├── README.md               # HF Spaces 配置（包含 sdk_version）
+├── database_setup_lite.py  # 数据库设置
+├── rag_engine.py           # RAG 引擎（已更新 max_tokens: 4000）
+└── mock_data.json          # 示例数据
+```
+
+## 🚀 部署步骤
+
+### 方法 1: 使用 Git（推荐）
+
+1. **克隆 HF Space 仓库**（如果还没有）：
+   ```bash
+   git clone https://huggingface.co/spaces/curio-lab/GraphRAG_Backend
+   cd GraphRAG_Backend
+   ```
+
+2. **复制所有文件到仓库根目录**：
+   ```bash
+   # 从 hf_deploy 目录复制所有文件
+   cp ../hf_deploy/* .
+   ```
+
+3. **提交并推送**：
+   ```bash
+   git add .
+   git commit -m "Update: Increase max_tokens to 4000"
+   git push origin main
+   ```
+
+### 方法 2: 使用 Web UI
+
+1. 访问 https://huggingface.co/spaces/curio-lab/GraphRAG_Backend
+2. 点击 **"Files and versions"** 标签
+3. 点击 **"Add file"** → **"Upload file"**
+4. 依次上传 `hf_deploy/` 目录下的所有文件：
+   - `app.py`
+   - `Dockerfile`
+   - `requirements.txt`
+   - `README.md`
+   - `database_setup_lite.py`
+   - `rag_engine.py`
+   - `mock_data.json`
+
+## ⚠️ 重要提示
+
+1. **文件必须在根目录**：所有文件必须上传到 HF Space 仓库的根目录，不要放在子文件夹中
+
+2. **环境变量**：确保在 HF Spaces Settings → Secrets 中设置了：
+   - `LLM_API_BASE` = `https://api.ai-gaochao.cn/v1`
+   - `LLM_API_KEY` = `sk-你的真实密钥`
+   - `LLM_MODEL` = `gemini-2.5-flash`
+   - `EMBEDDING_MODEL` = `text-embedding-3-small`
+
+3. **等待构建**：推送后，HF Spaces 会自动构建，通常需要 5-10 分钟
+
+## 📝 本次更新内容
+
+- ✅ `rag_engine.py`: 将 `max_tokens` 从 2000 增加到 4000，避免内容被截断
+- ✅ 所有文件已更新到最新版本
+

Dockerfile

@@ -0,0 +1,30 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# 安装系统依赖
+RUN apt-get update && apt-get install -y \
+    git \
+    && rm -rf /var/lib/apt/lists/*
+
+# 复制 requirements.txt
+COPY requirements.txt .
+
+# 安装 Python 依赖
+RUN pip install --no-cache-dir -r requirements.txt
+
+# 复制应用代码和依赖文件
+COPY app.py .
+COPY database_setup_lite.py .
+COPY rag_engine.py .
+COPY mock_data.json .
+
+# 暴露端口（HF Spaces 使用 7860）
+EXPOSE 7860
+
+# 设置环境变量
+ENV PORT=7860
+ENV TOKENIZERS_PARALLELISM=false
+
+# 启动应用（使用 uvicorn）
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -0,0 +1,35 @@
+---
+title: GraphRAG Backend
+emoji: 🕸️
+colorFrom: purple
+colorTo: pink
+sdk: docker
+sdk_version: "4.0.0"
+app_file: app.py
+pinned: false
+---
+
+# GraphRAG Backend API
+
+GraphRAG 后端服务，提供图增强检索和文案生成功能。
+
+## API 端点
+
+- `GET /` - API 信息
+- `GET /api/products` - 获取产品列表
+- `GET /api/styles` - 获取风格列表
+- `GET /api/graph` - 获取图数据
+- `GET /api/vector-db` - 获取向量数据库
+- `POST /api/search` - 对比检索
+- `POST /api/generate` - 生成文案
+- `POST /api/features/search` - 搜索特征
+
+## 环境变量
+
+在 Hugging Face Spaces 的 Settings → Secrets 中设置：
+
+- `LLM_API_BASE` - AI API 基础地址
+- `LLM_API_KEY` - AI API 密钥（敏感）
+- `LLM_MODEL` - AI 模型名称（可选）
+- `EMBEDDING_MODEL` - Embedding 模型名称（可选）
+

app.py

@@ -0,0 +1,203 @@
+"""
+FastAPI 后端服务 - 用于 Hugging Face Spaces
+"""
+import os
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+
+from fastapi import FastAPI, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+from typing import Optional, List
+import json
+
+# 导入数据库和 RAG 引擎
+# 注意：在 HF Spaces 中，这些文件应该在同一个目录下
+from database_setup_lite import setup_databases
+from rag_engine import RAGEngine
+
+# 初始化 FastAPI 应用
+app = FastAPI(title="GraphRAG Backend API")
+
+# 配置 CORS - 允许所有来源（生产环境可以限制为特定域名）
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],  # 生产环境可以设置为 ["https://your-frontend.vercel.app"]
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# 初始化数据库和引擎（全局变量，避免重复初始化）
+print("正在初始化数据库...")
+graph_db, vector_db = setup_databases()
+rag_engine = RAGEngine(graph_db, vector_db)
+
+# 加载数据用于前端展示
+with open("mock_data.json", "r", encoding="utf-8") as f:
+    mock_data = json.load(f)
+
+# Pydantic 模型
+class SearchRequest(BaseModel):
+    query: str
+    product_name: Optional[str] = ""
+    style_name: Optional[str] = ""
+
+class GenerateRequest(BaseModel):
+    query: str
+    product_name: str
+    style_name: str
+    use_graph: bool = True
+
+class FeatureSearchRequest(BaseModel):
+    query: str
+
+@app.get("/")
+def root():
+    """根路径"""
+    return {
+        "message": "GraphRAG Backend API",
+        "version": "1.0.0",
+        "endpoints": [
+            "GET /api/products",
+            "GET /api/styles",
+            "GET /api/graph",
+            "GET /api/vector-db",
+            "POST /api/search",
+            "POST /api/generate",
+            "POST /api/features/search"
+        ]
+    }
+
+@app.get("/api/products")
+def get_products():
+    """获取产品列表"""
+    demo_product = {
+        "id": "P_DEMO",
+        "name": "真丝睡眠眼罩"
+    }
+    return [demo_product]
+
+@app.get("/api/styles")
+def get_styles():
+    """获取风格列表"""
+    styles = [{"id": s["id"], "name": s["name"]} for s in mock_data["styles"]]
+    return styles
+
+@app.get("/api/graph")
+def get_graph():
+    """获取图结构数据"""
+    nodes = []
+    edges = []
+    
+    # 添加节点
+    for node_id, node_data in graph_db.nodes.items():
+        nodes.append({
+            "id": node_id,
+            "type": node_data["type"],
+            "label": node_data["properties"].get("name") or node_data["properties"].get("content", "")[:20] or node_id,
+            "properties": node_data["properties"]
+        })
+    
+    # 添加边
+    for edge in graph_db.edges:
+        edges.append({
+            "source": edge["source"],
+            "target": edge["target"],
+            "relationship": edge["relationship"]
+        })
+    
+    return {
+        "nodes": nodes,
+        "edges": edges
+    }
+
+@app.post("/api/search")
+def search(request: SearchRequest):
+    """搜索接口"""
+    if not request.query:
+        raise HTTPException(status_code=400, detail="查询不能为空")
+    
+    comparison = rag_engine.compare_retrieval(
+        request.query,
+        request.product_name or "",
+        request.style_name or ""
+    )
+    
+    return comparison
+
+@app.post("/api/generate")
+def generate(request: GenerateRequest):
+    """生成文案接口"""
+    if not all([request.query, request.product_name, request.style_name]):
+        raise HTTPException(status_code=400, detail="缺少必要参数")
+    
+    result = rag_engine.generate_copywriting(
+        request.query,
+        request.product_name,
+        request.style_name,
+        request.use_graph
+    )
+    
+    return result
+
+@app.get("/api/vector-db")
+def get_vector_db():
+    """获取传统RAG的向量数据库内容"""
+    try:
+        collection = vector_db.collection
+        all_docs = collection.get()
+        
+        documents = []
+        for i, doc_id in enumerate(all_docs["ids"]):
+            documents.append({
+                "id": doc_id,
+                "content": all_docs["documents"][i] if "documents" in all_docs and i < len(all_docs["documents"]) else "",
+                "metadata": all_docs["metadatas"][i] if "metadatas" in all_docs and i < len(all_docs["metadatas"]) else {}
+            })
+        
+        return {
+            "total": len(documents),
+            "documents": documents
+        }
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+@app.post("/api/features/search")
+def search_features(request: FeatureSearchRequest):
+    """根据查询搜索相关特征"""
+    query = request.query.lower()
+    
+    if not query:
+        return {"features": []}
+    
+    # 获取所有特征节点
+    feature_nodes = graph_db.find_nodes_by_type("Feature")
+    matched_features = []
+    
+    for node in feature_nodes:
+        feature_name = node["properties"].get("name", node["id"]).lower()
+        # 简单的关键词匹配
+        if query in feature_name or any(keyword in feature_name for keyword in query.split()):
+            matched_features.append({
+                "id": node["id"],
+                "name": node["properties"].get("name", node["id"]),
+                "related_products": []
+            })
+            
+            # 查找使用该特征的产品
+            for edge in graph_db.edges:
+                if edge["target"] == node["id"] and edge["relationship"] == "HAS_FEATURE":
+                    product_node = graph_db.nodes.get(edge["source"], {})
+                    if product_node.get("type") == "Product":
+                        matched_features[-1]["related_products"].append(
+                            product_node["properties"].get("name", edge["source"])
+                        )
+    
+    return {"features": matched_features[:10]}  # 最多返回10个
+
+# HF Spaces 会自动使用 Dockerfile 中的 CMD 启动
+# 如果需要本地测试，可以取消下面的注释
+# if __name__ == "__main__":
+#     import uvicorn
+#     port = int(os.getenv("PORT", 7860))
+#     uvicorn.run(app, host="0.0.0.0", port=port)

database_setup_lite.py

@@ -0,0 +1,325 @@
+"""
+轻量级图向量数据库 - 不依赖 ChromaDB，避免超过 Vercel 250MB 限制
+使用纯 Python 实现简单的向量搜索
+"""
+import os
+import json
+import math
+from typing import List, Dict, Optional
+import requests
+
+# 延迟导入 sentence_transformers，避免依赖冲突
+HAS_SENTENCE_TRANSFORMERS = False
+SentenceTransformer = None
+
+def _try_import_sentence_transformers():
+    """尝试导入 sentence_transformers"""
+    global HAS_SENTENCE_TRANSFORMERS, SentenceTransformer
+    if HAS_SENTENCE_TRANSFORMERS:
+        return True
+    try:
+        from sentence_transformers import SentenceTransformer as ST
+        SentenceTransformer = ST
+        HAS_SENTENCE_TRANSFORMERS = True
+        return True
+    except (ImportError, RuntimeError, AttributeError) as e:
+        HAS_SENTENCE_TRANSFORMERS = False
+        return False
+
+# 简单的内存图数据库
+class SimpleGraphDB:
+    """简单的内存图数据库模拟"""
+    def __init__(self):
+        self.nodes = {}  # {node_id: {type, properties}}
+        self.edges = []  # [{source, target, relationship}]
+    
+    def add_node(self, node_id: str, node_type: str, properties: Dict):
+        """添加节点"""
+        self.nodes[node_id] = {
+            "type": node_type,
+            "properties": properties
+        }
+    
+    def add_edge(self, source: str, target: str, relationship: str):
+        """添加边"""
+        self.edges.append({
+            "source": source,
+            "target": target,
+            "relationship": relationship
+        })
+    
+    def get_neighbors(self, node_id: str, relationship: Optional[str] = None) -> List[Dict]:
+        """获取邻居节点"""
+        neighbors = []
+        for edge in self.edges:
+            if edge["source"] == node_id:
+                if relationship is None or edge["relationship"] == relationship:
+                    target_node = self.nodes.get(edge["target"], {})
+                    neighbors.append({
+                        "node_id": edge["target"],
+                        "relationship": edge["relationship"],
+                        "properties": target_node.get("properties", {})
+                    })
+        return neighbors
+    
+    def find_nodes_by_type(self, node_type: str) -> List[Dict]:
+        """根据类型查找节点"""
+        return [
+            {"id": node_id, **node_data}
+            for node_id, node_data in self.nodes.items()
+            if node_data["type"] == node_type
+        ]
+    
+    def find_node_by_property(self, node_type: str, property_name: str, property_value: str) -> Optional[Dict]:
+        """根据属性查找节点"""
+        for node_id, node_data in self.nodes.items():
+            if node_data["type"] == node_type:
+                props = node_data.get("properties", {})
+                if props.get(property_name) == property_value:
+                    return {"id": node_id, **node_data}
+        return None
+
+def cosine_similarity(vec1: List[float], vec2: List[float]) -> float:
+    """计算余弦相似度"""
+    dot_product = sum(a * b for a, b in zip(vec1, vec2))
+    magnitude1 = math.sqrt(sum(a * a for a in vec1))
+    magnitude2 = math.sqrt(sum(a * a for a in vec2))
+    if magnitude1 == 0 or magnitude2 == 0:
+        return 0.0
+    return dot_product / (magnitude1 * magnitude2)
+
+class VectorDB:
+    """轻量级向量数据库 - 使用内存存储，不依赖 ChromaDB"""
+    def __init__(self):
+        # 文档存储：{id: {content, metadata, embedding}}
+        self.documents: Dict[str, Dict] = {}
+        
+        # Embedding 配置
+        self.embedding_api_base = os.getenv("LLM_API_BASE", "https://api.ai-gaochao.cn/v1")
+        self.embedding_api_key = os.getenv("LLM_API_KEY", "")
+        self.embedding_model = os.getenv("EMBEDDING_MODEL", "text-embedding-3-small")
+        self.use_openai_embedding = bool(self.embedding_api_key)
+        
+        if self.use_openai_embedding:
+            print(f"✅ 使用 OpenAI Embeddings API: {self.embedding_model}")
+        else:
+            print("ℹ️  使用简单文本匹配（关键词搜索）")
+    
+    def _get_openai_embeddings(self, texts: List[str]) -> List[List[float]]:
+        """调用 OpenAI Embeddings API 获取向量"""
+        url = f"{self.embedding_api_base}/embeddings"
+        headers = {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {self.embedding_api_key}"
+        }
+        data = {
+            "input": texts,
+            "model": self.embedding_model
+        }
+        
+        response = requests.post(url, headers=headers, json=data, timeout=30)
+        response.raise_for_status()
+        result = response.json()
+        
+        return [item["embedding"] for item in result["data"]]
+    
+    def _simple_text_match(self, query: str, document: str) -> float:
+        """简单的文本匹配评分（关键词匹配）"""
+        query_words = set(query.lower().split())
+        doc_words = set(document.lower().split())
+        
+        if not query_words:
+            return 0.0
+        
+        # 计算匹配的关键词比例
+        matches = len(query_words & doc_words)
+        return matches / len(query_words)
+    
+    def add_documents(self, documents: List[str], ids: List[str], metadatas: List[Dict]):
+        """添加文档到向量数据库"""
+        if self.use_openai_embedding:
+            # 使用 OpenAI Embeddings API
+            try:
+                embeddings = self._get_openai_embeddings(documents)
+                for doc, doc_id, meta, emb in zip(documents, ids, metadatas, embeddings):
+                    self.documents[doc_id] = {
+                        "content": doc,
+                        "metadata": meta,
+                        "embedding": emb
+                    }
+            except Exception as e:
+                print(f"⚠️  OpenAI Embeddings API 调用失败: {e}")
+                # 回退到简单存储（无 embedding）
+                for doc, doc_id, meta in zip(documents, ids, metadatas):
+                    self.documents[doc_id] = {
+                        "content": doc,
+                        "metadata": meta,
+                        "embedding": None
+                    }
+        else:
+            # 不使用 embedding，只存储文档
+            for doc, doc_id, meta in zip(documents, ids, metadatas):
+                self.documents[doc_id] = {
+                    "content": doc,
+                    "metadata": meta,
+                    "embedding": None
+                }
+    
+    def search(self, query: str, n_results: int = 5) -> List[Dict]:
+        """语义搜索"""
+        if self.use_openai_embedding:
+            # 使用向量相似度搜索
+            try:
+                query_embedding = self._get_openai_embeddings([query])[0]
+                
+                # 计算所有文档的相似度
+                results = []
+                for doc_id, doc_data in self.documents.items():
+                    if doc_data["embedding"]:
+                        similarity = cosine_similarity(query_embedding, doc_data["embedding"])
+                        results.append({
+                            "content": doc_data["content"],
+                            "metadata": doc_data["metadata"],
+                            "distance": 1 - similarity,  # 转换为距离（越小越相似）
+                            "id": doc_id
+                        })
+                
+                # 按相似度排序
+                results.sort(key=lambda x: x["distance"])
+                return results[:n_results]
+            except Exception as e:
+                print(f"⚠️  向量搜索失败，回退到文本匹配: {e}")
+                # 回退到文本匹配
+                return self._text_search(query, n_results)
+        else:
+            # 使用简单文本匹配
+            return self._text_search(query, n_results)
+    
+    def _text_search(self, query: str, n_results: int) -> List[Dict]:
+        """简单的文本匹配搜索"""
+        results = []
+        for doc_id, doc_data in self.documents.items():
+            score = self._simple_text_match(query, doc_data["content"])
+            if score > 0:
+                results.append({
+                    "content": doc_data["content"],
+                    "metadata": doc_data["metadata"],
+                    "distance": 1 - score,  # 转换为距离
+                    "id": doc_id
+                })
+        
+        # 按相似度排序
+        results.sort(key=lambda x: x["distance"])
+        return results[:n_results]
+    
+    @property
+    def collection(self):
+        """兼容性属性，模拟 ChromaDB 的 collection 接口"""
+        class MockCollection:
+            def __init__(self, vector_db):
+                self.vector_db = vector_db
+            
+            def get(self):
+                """获取所有文档"""
+                ids = list(self.vector_db.documents.keys())
+                documents = [self.vector_db.documents[id]["content"] for id in ids]
+                metadatas = [self.vector_db.documents[id]["metadata"] for id in ids]
+                return {
+                    "ids": ids,
+                    "documents": documents,
+                    "metadatas": metadatas
+                }
+        
+        return MockCollection(self)
+
+def setup_databases():
+    """初始化数据库"""
+    # 加载数据
+    with open("mock_data.json", "r", encoding="utf-8") as f:
+        data = json.load(f)
+    
+    # 初始化图数据库
+    graph_db = SimpleGraphDB()
+    
+    # 添加产品节点
+    for product in data["products"]:
+        graph_db.add_node(
+            product["id"],
+            "Product",
+            {
+                "name": product["name"],
+                "type": product["type"],
+                "keywords": product["keywords"],
+                "features": product["features"]
+            }
+        )
+    
+    # 添加风格节点
+    for style in data["styles"]:
+        graph_db.add_node(
+            style["id"],
+            "Style",
+            {
+                "name": style["name"],
+                "description": style["description"],
+                "characteristics": style["characteristics"]
+            }
+        )
+    
+    # 添加文案节点
+    for copy in data["copywritings"]:
+        graph_db.add_node(
+            copy["id"],
+            "Copywriting",
+            {
+                "content": copy["content"],
+                "tag": copy["tag"],
+                "target_audience": copy["target_audience"]
+            }
+        )
+    
+    # 添加特征节点
+    all_features = set()
+    for product in data["products"]:
+        for feature in product.get("features", []):
+            all_features.add(feature)
+    
+    for feature in all_features:
+        graph_db.add_node(feature, "Feature", {"name": feature})
+    
+    # 添加关系
+    for rel in data["relationships"]:
+        graph_db.add_edge(
+            rel["source"],
+            rel["target"],
+            rel["relationship"]
+        )
+    
+    # 初始化轻量级向量数据库
+    vector_db = VectorDB()
+    
+    # 添加文案到向量数据库
+    documents = []
+    ids = []
+    metadatas = []
+    
+    for copy in data["copywritings"]:
+        documents.append(copy["content"])
+        ids.append(copy["id"])
+        metadatas.append({
+            "product_id": copy["product_id"],
+            "style_id": copy["style_id"],
+            "tag": copy["tag"],
+            "target_audience": copy["target_audience"]
+        })
+    
+    vector_db.add_documents(documents, ids, metadatas)
+    print(f"✅ 向量数据库已更新，包含 {len(documents)} 个文案")
+    
+    print("数据库初始化完成！")
+    print(f"- 图数据库节点数: {len(graph_db.nodes)}")
+    print(f"- 图数据库边数: {len(graph_db.edges)}")
+    print(f"- 向量数据库文档数: {len(documents)}")
+    
+    return graph_db, vector_db
+

mock_data.json

@@ -0,0 +1,1108 @@
+{
+  "products": [
+    {
+      "id": "P1",
+      "name": "索尼降噪耳机 WH-1000XM5",
+      "type": "数码",
+      "keywords": [
+        "降噪",
+        "隔绝世界",
+        "静谧",
+        "安静",
+        "独处"
+      ],
+      "features": [
+        "主动降噪",
+        "长续航",
+        "舒适佩戴",
+        "便携",
+        "静音设计"
+      ]
+    },
+    {
+      "id": "P2",
+      "name": "祖玛珑香薰蜡烛 英国梨与小苍兰",
+      "type": "家居",
+      "keywords": [
+        "独处",
+        "氛围感",
+        "治愈",
+        "安静",
+        "避难所"
+      ],
+      "features": [
+        "持久留香",
+        "天然蜡质",
+        "精致包装",
+        "环保材质"
+      ]
+    },
+    {
+      "id": "P3",
+      "name": "真丝睡眠眼罩",
+      "type": "家居",
+      "keywords": [
+        "遮光",
+        "安静",
+        "独处",
+        "避难所",
+        "断联"
+      ],
+      "features": [
+        "100%真丝",
+        "完全遮光",
+        "舒适贴合",
+        "便携"
+      ]
+    },
+    {
+      "id": "P4",
+      "name": "复古CCD相机",
+      "type": "数码",
+      "keywords": [
+        "颗粒感",
+        "低像素",
+        "复古",
+        "怀旧",
+        "氛围感"
+      ],
+      "features": [
+        "复古外观",
+        "颗粒感照片",
+        "轻便便携",
+        "便携"
+      ]
+    },
+    {
+      "id": "P5",
+      "name": "手冲咖啡套装",
+      "type": "生活",
+      "keywords": [
+        "仪式感",
+        "慢生活",
+        "独处",
+        "治愈"
+      ],
+      "features": [
+        "精品咖啡豆",
+        "专业器具",
+        "教程指导",
+        "精致包装"
+      ]
+    },
+    {
+      "id": "P6",
+      "name": "无印良品香薰机",
+      "type": "家居",
+      "keywords": [
+        "氛围感",
+        "治愈",
+        "安静",
+        "独处",
+        "放松"
+      ],
+      "features": [
+        "静音设计",
+        "LED夜灯",
+        "定时功能"
+      ]
+    },
+    {
+      "id": "P7",
+      "name": "Kindle电子书阅读器",
+      "type": "数码",
+      "keywords": [
+        "安静",
+        "独处",
+        "思考",
+        "避世",
+        "自我"
+      ],
+      "features": [
+        "护眼屏幕",
+        "长续航",
+        "海量图书",
+        "便携"
+      ]
+    },
+    {
+      "id": "P8",
+      "name": "瑜伽垫",
+      "type": "运动",
+      "keywords": [
+        "放松",
+        "治愈",
+        "独处",
+        "自我",
+        "安静"
+      ],
+      "features": [
+        "防滑设计",
+        "环保材质",
+        "便携",
+        "舒适贴合"
+      ]
+    },
+    {
+      "id": "P9",
+      "name": "蓝牙音箱",
+      "type": "数码",
+      "keywords": [
+        "氛围感",
+        "治愈",
+        "放松",
+        "享受"
+      ],
+      "features": [
+        "360度环绕",
+        "长续航",
+        "防水设计"
+      ]
+    },
+    {
+      "id": "P10",
+      "name": "茶具套装",
+      "type": "生活",
+      "keywords": [
+        "仪式感",
+        "慢生活",
+        "独处",
+        "思考"
+      ],
+      "features": [
+        "精美设计",
+        "陶瓷材质",
+        "礼盒包装",
+        "精致包装",
+        "环保材质"
+      ]
+    }
+  ],
+  "styles": [
+    {
+      "id": "S1",
+      "name": "清冷避世风",
+      "description": "强调孤独、逃离人群、自我对话",
+      "characteristics": [
+        "孤独",
+        "避世",
+        "自我",
+        "安静",
+        "独处",
+        "避难所"
+      ]
+    },
+    {
+      "id": "S2",
+      "name": "疯狂种草风",
+      "description": "强调断货、颜值、必须买",
+      "characteristics": [
+        "绝绝子",
+        "必买",
+        "断货",
+        "颜值",
+        "种草"
+      ]
+    },
+    {
+      "id": "S3",
+      "name": "深夜emo风",
+      "description": "深夜情绪、感伤、思考",
+      "characteristics": [
+        "深夜",
+        "emo",
+        "感伤",
+        "思考",
+        "情绪"
+      ]
+    },
+    {
+      "id": "S4",
+      "name": "治愈系风格",
+      "description": "温暖、治愈、放松",
+      "characteristics": [
+        "治愈",
+        "温暖",
+        "放松",
+        "舒适",
+        "安心"
+      ]
+    },
+    {
+      "id": "S5",
+      "name": "极简生活风",
+      "description": "简约、品质、生活态度",
+      "characteristics": [
+        "简约",
+        "品质",
+        "生活",
+        "态度",
+        "精致"
+      ]
+    }
+  ],
+  "copywritings": [
+    {
+      "id": "C1",
+      "product_id": "P2",
+      "style_id": "S1",
+      "content": "夜幕低垂，城市的喧嚣仿佛被一层无形的结界隔绝，此刻，我只想退守一隅，与自己对话。点燃祖玛珑英国梨与小苍兰香薰蜡���，那一缕清冽而又微甜的香气，便如薄雾般温柔弥漫开来，瞬间构建起我的专属“避难所”。\n\n它不张扬，不聒噪，是恰到好处的疏离感。前调的英国梨带来一抹清新的果香，不甜腻，反而透着一丝清冷，恰似秋日清晨的露珠。随之而来的小苍兰，则赋予空间一抹温柔的芬芳，不浓烈，却足够治愈。这香气是持久留香的，无需刻意捕捉，它便如影随形，伴我度过每一个沉静的夜晚。\n\n我喜欢这种与世无争的氛围，它让我在独处中找到真正的自我。窝在沙发一角，翻开一本旧书，或只是静静地冥想，任由思绪在香气的引导下自由飘荡。纯粹的天然蜡质，燃烧时没有一丝杂质，只有干净的火焰和纯粹的香气，这与我对生活极简主义的追求不谋而合。精致的磨砂玻璃瓶身，无需点燃时也自成一道风景，低调而有格调，彰显着一种无声的品味。\n\n它不仅仅是一支蜡烛，更是我远离尘嚣的仪式感，是喧嚣世界里一片私密的安静角落。在这片小小的避难所里，我得以疗愈身心，与内心深处的自我坦诚相见。",
+      "tag": "HOT",
+      "target_audience": "都市独居青年"
+    },
+    {
+      "id": "C2",
+      "product_id": "P2",
+      "style_id": "S2",
+      "content": "绝绝子姐妹们！我宣布！祖玛珑英国梨与小苍兰香薰蜡烛，绝对是你们独处时光的灵魂伴侣！😭 香氛控们听好了，这支根本就是艺术品，颜值高到爆炸，摆在哪都高级感满满，简直是家居氛围感的杀手锏！\n\n想象一下，忙碌一天后，卸下疲惫，回到只属于你的小天地。点燃它，瞬间，整个房间都被那温柔又高级的香气包裹。前调是清新的英国梨，甜而不腻，带着露水的湿润感，随后小苍兰的优雅与含蓄缓缓绽放，最后以广藿香的沉稳收尾。这种多层次的治愈香气，真的能让人瞬间放松下来，仿佛置身于宁静的英式花园，忘却所有烦恼，找到属于自己的那份安静避难所。\n\n它不仅留香超级持久，天然蜡质燃烧起来也非常干净，环保材质也让人用得安心。每一次点燃，那温暖的光晕和弥漫的香气，都让我的独处时间变得格外有仪式感。它不是简单的香薰，更是一种生活情调，一种对自己温柔的投资！\n\n实话告诉你们，这支香薰蜡烛真的是“断货王”中的“断货王”！每次补货都秒空，根本抢不到！所以姐妹们，如果你看到它，请务必、立刻、马上抱走！这是我年度必买清单No.1，不买真的会后悔一整年！快冲啊！💖",
+      "tag": "HOT",
+      "target_audience": "年轻女性"
+    },
+    {
+      "id": "C3",
+      "product_id": "P2",
+      "style_id": "S3",
+      "content": "夜深了，窗外细雨沥沥，世界仿佛沉入一片寂静。是不是每个人，都有那么几个瞬间，渴望把自己包裹起来，寻一个无人打扰的“避难所”？当那些无处安放的思绪涌上心头，感伤和思考交织，我总会习惯性地走向它——祖玛珑英国梨与小苍兰香薰蜡烛。\n\n轻柔地点燃，微弱的火光在昏暗中跳动，瞬间，整个空间都被那股独特的香气温柔地笼罩。它不是那种咄咄逼人的浓郁，而更像是一场细腻的耳语：英国梨的清甜带着一丝恰到好处的凉意，小苍兰的优雅则如同月光般倾泻而下，缓缓抚平了白天所有的喧嚣和疲惫。这份香气持久留香，一点点渗入空气，也一点点渗入我的心底。\n\n我喜欢在这样的深夜，放下手机，任由天然蜡质燃烧时那份纯净的气息，与我此刻的情绪融为一体。烛光把我的影子拉得很长很长，看着它精致的包装，触碰着环保材质的烛杯，所有的细节都在无声地诉说着一份对生活、对自我的温柔呵护。这份独处的“氛围感”被极致放大，让感伤的情绪不再那么尖锐，反而多了一份沉淀后的安宁。\n\n它不仅仅是一支香薰蜡烛，更像是我深夜里那位无声的知己。它不评判，不打扰，只是默默地用它温柔的光和香，为我搭建起一个安静的港湾，一个可以自由emo、思考人生的专属空间。在这里，心底那些纷繁复杂的情绪，似乎都被这治愈的香气温柔地安抚。那一刻，我与香气共鸣，与孤独和解，在光影交错中，找回了久违的平静与力量。",
+      "tag": "HOT",
+      "target_audience": "深夜思考者"
+    },
+    {
+      "id": "C4",
+      "product_id": "P1",
+      "style_id": "S1",
+      "content": "在都市的喧嚣中，你是否也渴望一处纯粹的空谷，只闻心跳，不染尘嚣？🎧 索尼WH-1000XM5，于我而言，早已超越了耳机的范畴，它是我的便携式“避难所”，是我与这个世界之间，那道若有似无的结界。\n\n戴上它的一瞬，世界仿佛按下静音键，所有冗余的嘈杂被温柔且坚定地隔���在外。咖啡馆的窃窃私语、地铁的轰鸣、甚至窗外的风声，都化作了背景里模糊的低语，不再有侵扰。这份极致的主动降噪，像为我的灵魂筑起了一方无声的穹顶，只留给我最纯粹的静谧。\n\n在这片由XM5构筑的独处空间里，我可以毫无保留地与自我对话。不必去迎合外界的期待，不用去理会人群的纷扰。我听得见自己的呼吸，感受得到思绪的流淌，每一次心跳都变得清晰而有力。它不只是隔绝噪音，更是提供了一个让内心得以沉淀、自我得以生长的避世之所。\n\n长达数十小时的续航，让这份清冷与自我，得以从晨曦延续至深夜，无需担忧被打断的烦恼。而轻若无物的舒适佩戴感，即便在漫长的冥想或深度阅读中，也感受不到丝毫压迫，仿佛它本就是你的一部分。它的静音设计，简约而深沉，与我的避世美学不谋而合。这不叫孤独，这叫在浮躁世界里，为自己觅得一隅安心的净土。它是我的“暂停键”，更是每一次心神",
+      "tag": "COLD",
+      "target_audience": "通勤族"
+    },
+    {
+      "id": "C5",
+      "product_id": "P1",
+      "style_id": "S2",
+      "content": "姐妹们！我真的挖到宝藏了！这款索尼WH-1000XM5降噪耳机，简直是我的救命稻草，直接刷新了我对“安静”的认知！这颜值，绝绝子！高级感直接拉满，戴上你就是氛围感女王本人！\n\n以前通勤地铁上人声鼎沸，办公室键盘声此起彼伏，想安静地看本书、听首歌都难。自从有了它，一键开启，整个世界瞬间“静音”！那种被温柔包裹的感觉，太治愈了！所有的喧嚣、嘈杂，统统被它隔绝在外，瞬间拥有一个完全属于自己的静谧小宇宙。它不是简单的降噪，而是给你一片极致的安静，让你能真正沉浸在自己的独处时光里。\n\n而且，它的佩戴舒适度简直无敌，轻若无物，戴一天耳朵都不会累，仿佛没戴一样。长续航更是出差旅行必备，告别电量焦虑，无论多长的旅途，它都能稳稳陪伴。便携性也超赞，轻巧收纳，不占空间，随时随地都能享受这份专属的静谧。\n\n真的，姐妹们！如果你也渴望一份属于自己的安静，想要把这个浮躁的世界“静音”，这款索尼XM5你真的不能错过！它已经不止是耳机了，它是你心灵的避风港，是你的私人静谧舱！这波不冲你一定会后悔！最近已经有断货趋势了，赶紧行动起来，不然真的抢不到啦！快去拥有它，给你的耳朵和心灵一个最温柔的拥抱吧！",
+      "tag": "HOT",
+      "target_audience": "数码爱好者"
+    },
+    {
+      "id": "C6",
+      "product_id": "P4",
+      "style_id": "S2",
+      "content": "姐妹们！救命🆘！我挖到宝藏了！这台复古CCD相机简直是我的梦中情机，颜值逆天，直接把我带回千禧年！ ✨\n\n自从入手了它，我的手机拍照功能都失宠了！它的外观是那种超有质感的复古设计，拿在手里就是时尚单品，随便一拍都是大片既视感，秒杀一切滤镜。但真正让我沦陷的，是它拍出来的照片！那独特的低像素颗粒感，简直绝绝子！不是那种模糊感，而是自带电影胶片滤镜的怀旧风情，每一张都充满故事和浓浓的氛围感，瞬间把日常场景变成艺术品。\n\n无论是和姐妹下午茶，还是周末出去Citywalk，或者记录旅行沿途的风景，它轻巧到可以随便塞进小包，完全没有负担。拍出来的照片不用P，自带的复古调调就高级到不行，那种时间沉淀下来的温暖感，让你的回忆都变得特别有温度。再也不用花大把时间调色找滤镜了，随手一按，就是氛围感天花板！\n\n我跟你们说，这玩意儿真的太火了，好多地方都断货了！如果你也爱复古风，追求那种独特的视觉体验，那么这台CCD相机你一定要拥有！它会让你重新爱上记录生活，感受每一个瞬间的美好，绝对是今年必入的单品，不买真的会后悔一辈子啊！",
+      "tag": "HOT",
+      "target_audience": "摄影爱好者"
+    },
+    {
+      "id": "C7",
+      "product_id": "P5",
+      "style_id": "S1",
+      "content": "当城市被喧嚣吞噬，夜色温柔地落下，我选择将自己抽离，遁入一个只属于我的安静角落。窗外霓虹闪烁，而我的世界，只剩下咖啡研磨的细碎声和水流轻柔地落在咖啡粉上的沙沙响。☕️ 这套手冲咖啡套装，不只是一组工具，它是我的避难所，是我与自我对话的个人仪式。\n\n打开精致的包装，里面是精心挑选的精品咖啡豆，带着泥土和阳光的气息，等待被唤醒。专业的器具冰冷而有温度，它们等待着被赋予生命。我依照教程指引，每一步都充满专注，细致地研磨，再用温热的水，以缓慢而均匀的速度，画着圈圈，让咖啡的灵魂一点点被唤醒���\n\n氤氲的香气，像一张无形的网，将我与尘嚣彻底隔开。此刻，没有社交的压力，没有琐碎的烦扰，只有我和我的咖啡，以及内心深处最真实的对话。这是一种孤独，却也无比治愈的独处。它让我慢下来，让思绪沉淀，每一口都带着深邃的苦涩和回甘，像极了人生，需要细细品味。\n\n它不是为了取悦任何人，而是为了滋养自我。在咖啡的暖意中，我找到了片刻的安宁，一个专属的安静角落。在这里，我可以卸下所有的伪装，与自我坦诚相待。这份清冷避世的慢生活，是我送给自己最好的礼物。✨",
+      "tag": "COLD",
+      "target_audience": "咖啡爱好者"
+    },
+    {
+      "id": "C8",
+      "product_id": "P6",
+      "style_id": "S4",
+      "content": "忙碌了一天，最期待的就是回到家，卸下所有疲惫，沉浸在只属于自己的小世界里。这时候，我的无印良品香薰机就是我的治愈法宝。它轻柔地吐出细密的雾气，带着我最爱的精油香气，瞬间让整个房间被温暖又安静的氛围感包裹。那种由内而外的放松和治愈，真的能让一天的烦恼都烟消云散。\n\n最爱它的静音设计，几乎感受不到它的存在，只有淡淡的香气和柔和的LED夜灯在默默陪伴。无论是深夜窝在沙发里读一本书，还是只是静静发呆，享受独处的安静时光，这份不被打扰的舒适感都弥足珍贵。开启定时功能，让香气温柔地伴随我缓缓进入甜美的梦乡，再也不用担心忘记关掉，安心感十足。\n\n看着它散发出的袅袅白雾，仿佛时间都慢了下来。它不仅仅是一台香薰机，更是我打造一方舒适安心小天地的秘密武器。在浮躁的世界里，给自己留一点与自我对话的时间，被温柔包围，感受生活中每一个值得珍藏的小确幸。这份温暖与舒适，值得每一个你拥有。",
+      "tag": "HOT",
+      "target_audience": "都市白领"
+    },
+    {
+      "id": "C9",
+      "product_id": "P6",
+      "style_id": "S1",
+      "content": "当世界按下静音键，只有我和无印良品香薰机。\n\n都市的喧嚣与疲惫，在踏入家门的那一刻，仿佛被一道无形的墙壁隔绝。我喜欢这种抽离感，将所有嘈杂抛诸脑后，只留下最纯粹的自我。这时，无印良品香薰机，成了我构筑这份避世清冷的唯一伴侣。\n\n它静默地立于一隅，没有一丝多余的色彩，如同我心境的投射。轻柔的水雾缓缓升腾，携带着我钟爱的木质香气，在空气中无声弥散。这不仅仅是“氛围感”，更是一种治愈系的结界，将我圈入一个只属于自己的小宇宙。静音设计是它的温柔，没有任何电机噪音能打破这份珍贵的安静，让我的思绪得以完全沉淀，与自我进行深层对话。\n\n夜幕降临，开启LED夜灯，那束微弱而温暖的光，如同夜空中最遥远的星辰，照亮我的阅读页，或者仅仅是陪伴我放空。不需要强烈的刺激，只需要这样淡淡的存在，就能带来极致的放松。定时功能更是贴心，它懂得我需要一个不被打扰的独处时光，也懂得何时让这一切温柔地结束。在这里，孤独并非寂寞，而是一种选择，一种力量。我的房间，我的避难所，因它而变得更加完整。在这里，我与世无争，只与自己同在。",
+      "tag": "HOT",
+      "target_audience": "都市独居青年"
+    },
+    {
+      "id": "C10",
+      "product_id": "P7",
+      "style_id": "S4",
+      "content": "最近是不是也觉得，生活节奏快到让人有点喘不过气？🌿 我找到了一个可以随时‘避世’的治愈小空间，那就是我的Kindle电子书阅读器。\n\n下班后，窝进柔软的沙发里，或是周末午后在阳光洒进的飘窗边，拿起它，世界的喧嚣仿佛瞬间被隔绝在外。那块特有的护眼屏幕，真的做到了像纸质书一样的温柔，长时间阅读眼睛也一点不觉得累，简直是为我们这些书虫量身定制的舒适感。📖\n\n最爱它超长续航的安心感，常常一两周都不用充电，让我可以完全沉浸在故事里，不必担心电量不足的焦虑，这种无打扰的体验太珍贵了。📚 海量的图书资源，无论是想探索浩瀚的宇宙，还是沉浸在浪漫的文学世界，都能轻松找到，每一次翻阅都是一次全新的自我发现之旅。\n\n小小一只，轻巧便携，无论是通勤路上，还是旅行途中，它都是那个默默陪伴你的知己。它不仅是冰冷的数码产品，更像是一个温暖的港湾，为你提供独处的机会，思考人生的真谛，找回内心深处的平静。这种只属于你和文字的亲密时刻，是现代生活中不可多得的治愈力量。✨ 拥有它，就拥有了一片属于自己的安静天地，让每一次阅读都成为一场灵魂的洗礼，温柔且坚定地与自我对话。",
+      "tag": "HOT",
+      "target_audience": "阅读爱好者"
+    },
+    {
+      "id": "C11",
+      "product_id": "P7",
+      "style_id": "S1",
+      "content": "当外界喧嚣渐远，当人群的浮躁令人疲惫，我需要一个只属于自己的角落。Kindle，便是我清冷的避难所，是我与世界保持距离的温柔屏障。它不是冰冷的数码设备，而是通往无数精神世界的密钥，是我与自我深层对话的唯一听众。\n\n指尖轻触，护眼屏幕上墨色文字温柔浮现，没有刺眼的光源，只有知识的温润与智慧的沉淀。我常在深夜，城市灯火阑珊时；在雨后的窗边，听雨打芭蕉声；或是在人迹罕至的山顶，呼吸清冽的空气，打开它。长续航让我的独处时光无需被外界琐事打扰，我可以放心地将整个下午或夜晚，沉浸在哲学巨著的浩瀚思考中，或在异世奇幻里避世长游，让思绪自由驰骋。\n\n海量图书是取之不尽的精神宝藏，从卡尔维诺的奇幻旅程到叔本华的孤独哲学，从村上春树的都市迷茫到梭罗的湖畔沉思，都能在此寻得共鸣。每一本书都是一个独立宇宙，而Kindle，就是通往这些宇宙的便携式飞船。它轻巧得仿佛不存在，却能装下整个宇宙的智慧与情感，让我的精神世界无远弗届。\n\n手握Kindle，我不是在简单的阅读，而是在进行一场私密的精神漫游。那些在书中相遇的灵魂，与我一同避开尘世的喧嚣，在文字构筑的宁静空间里，重新审视自我，沉淀思绪。孤独是我的勋章，独处是我的享受。Kindle，予我一片宁静的栖息地，让我安心地与自己的",
+      "tag": "HOT",
+      "target_audience": "通勤族"
+    },
+    {
+      "id": "C12",
+      "product_id": "P8",
+      "style_id": "S4",
+      "content": "姐妹们，有没有觉得最近生活节奏太快，身心俱疲？我最近爱上了在家里的独处瑜伽时光，真的是太治愈了！今天想跟大家分享我的秘密武器——这款让我爱不释手的瑜伽垫。\n\n每次下班回家，褪去一天的疲惫，铺开这张瑜伽垫，就像是开启了一个只属于我的安静结界。它的环保材质摸起来特别亲肤，没有任何异味，让我能安心地进行深呼吸，感觉就像是与大自然亲密接触。踩上去的那一刻，感觉整个人都被温柔地包裹起来，软硬适中，不会陷进去，也不会硬邦邦，完美地舒适贴合着身体曲线。\n\n最让我惊喜的是它的防滑设计！无论是做猫牛式舒展脊柱，还是挑战一些平衡体式，它都能稳稳地抓住地面，给我满满的安全感，再也不用担心滑倒啦。而且，它的舒适贴合度真的超赞，每一次伸展，每一次扭转，都能感觉到身体与垫子完美融合，动作也变得更加流畅自如。\n\n周末的清晨，泡一杯热茶，放上轻柔的冥想音乐，在这张瑜伽垫上练习，感受身体的苏醒与放松，仿佛所有的压力和烦恼都随着汗水一点点释放，找回内心的平衡。它不仅仅是一块运动垫，更是我寻找内心平静，实现自我治愈的小天地。因为它便携的设计，我还经常带它去公园，享受阳光下的瑜伽，把那份治愈带到户外。\n\n如果你也渴望拥有这样一份独处的治愈时光，真的推荐你试试它。让身体和心灵都能在这份柔软与安静中，找到久违的放松与平衡，体验真正的自我治愈。",
+      "tag": "HOT",
+      "target_audience": "瑜伽爱好者"
+    },
+    {
+      "id": "C13",
+      "product_id": "P8",
+      "style_id": "S1",
+      "content": "城市万象喧嚣，人群川流不息，总有那么一刻，只想逃离，寻一隅清净，与自我深谈。🧘‍♀️ 我的“避难所”，就是这张看似普通的瑜伽垫。当它轻柔地铺展开来，便瞬间隔绝了世界的芜杂，为我圈出一片独享的宁静天地。\n\n指尖触及的，是来自环保材质的纯粹与温柔，冰凉中带着一丝安抚。脚下的防滑设计，稳稳地托住我的每寸肌肤，让我不再担心外界的跌宕，只专注于内心的平衡与呼吸的节奏。✨ 在这里，没有旁人的目光，没有催促的声音，只有我自己，和这片专属的方寸之地。每一次缓慢的舒展，每一次沉静的冥想，都是与“自我”最深度、最真实的对话，仿佛在安静的宇宙里，独自漂浮，没有束缚。\n\n这种清冷而纯粹的独处，像一剂良药，治愈着日常积累的疲惫与浮躁。它不只是一件运动器械，更是我心灵的避难所，一个只属于我的精神港湾。当外界嘈杂入耳，我便会自然而然地走向它，寻求那份不言而喻的安静。而且它的轻巧便携，让我可以将这份宁静随身携带，无论是窗边、阳台，还是旅行中的一角，只要有它，我便能随时进入我的“避世”模式，找到那份舒适贴合的安稳。\n\n寻一个专属的角落，让身心在这片宁静中得到最深沉的放松与疗愈。这不是逃避，���是为了更好地面对，在孤独中找到力量与安宁。🌿",
+      "tag": "HOT",
+      "target_audience": "瑜伽爱好者"
+    },
+    {
+      "id": "C14",
+      "product_id": "P9",
+      "style_id": "S2",
+      "content": "OMG！姐妹们！最近挖到个蓝牙音箱绝绝子！氛围感秒拉满，不买会后悔一辈子！\n\n真的要疯狂安利这个宝藏音箱！颜值高到犯规，高级感满满，摆在哪儿都像一件艺术品，瞬间提升家居品味！它的360度环绕音效简直不要",
+      "tag": "HOT",
+      "target_audience": "音乐爱好者"
+    },
+    {
+      "id": "C15",
+      "product_id": "P9",
+      "style_id": "S3",
+      "content": "夜深了，城市万籁俱寂，只有指尖轻触屏幕的微光，和心底那一丝挥之不去的emo。耳机总觉得隔绝了世界，手机外放又少了点仪式感。直到我遇见了这个小小的蓝牙音箱，它成了我深夜情绪的专属治愈师。\n\n当熟悉的旋律从音箱中缓缓流淌出来，360度环绕音效瞬间将我包裹，不再是单薄的音符，而是像一层温柔的薄雾，将我笼罩在一个只属于我的声音结界里。闭上眼，仿佛置身于某个遥远而宁静的海岸，海浪声、风声、歌声，所有的一切都变得立体而真实。那些白日里堆积的疲惫、委屈和迷茫，在这一刻，都被音乐轻轻抚平，心里涌动的感伤，也仿佛找到了宣泄的出口，得到了温柔的安抚。\n\n它的长续航能力也让人格外安心，深夜的思绪总是漫长而无止境，无需担心电量耗尽会打断这份珍贵的沉浸。我可以任由自己沉溺在音乐的海洋中，从星辰满天到晨曦微露。更惊喜的是，它还拥有防水设计。有时，我会带着它走进浴室，让温热的水汽与舒缓的音乐一同蒸腾，洗去一天的尘埃和烦恼。在水声和歌声的交织中，我允许自己放下所有防备，享受这份极致的放松。\n\n它不只是一款音箱，更是深夜里一份懂得你心事的温柔陪伴。它带来的不止是声音，更是一种无法言喻的「氛围感」，一种深沉的「治愈」力量，以及最纯粹的「放松」与「享受」。深夜emo，有它在，好像也没那么孤独了。",
+      "tag": "HOT",
+      "target_audience": "深夜思考者"
+    },
+    {
+      "id": "C16",
+      "product_id": "P10",
+      "style_id": "S5",
+      "content": "都市喧嚣里，你是否也渴望一方宁静，一份属于自己的慢时光？今天想和大家分享一套让我爱不释手的【极简陶瓷茶具套装】，它不仅仅是一套茶具，更是我开启品质慢生活的仪式感入口。\n\n精美的设计，线条流畅，没有任何多余的装饰，完美诠释了“less is more”的极简美学。陶瓷材质温润如玉，握在手中，指尖能感受到细腻的触感，每一杯茶都仿佛被赋予了温度。更值得一提的是，它选用环保材质，在享受生活的同时，也对地球多了一份温柔。\n\n清晨，阳光透过窗帘洒进屋子，泡一壶清茶，任由茶香氤氲。一个人的午后，它陪伴我阅读、思考，或是放空，让思绪在茶烟袅袅中自由飞舞。这份独处的宁静，是都市人最奢侈的享受，它教会我如何在忙碌中找到平衡，重拾生活的节奏感。\n\n每次用它泡茶，都像在进行一场小小的仪式",
+      "tag": "HOT",
+      "target_audience": "茶文化爱好者"
+    },
+    {
+      "id": "C17",
+      "product_id": "P10",
+      "style_id": "S1",
+      "content": "城市喧嚣渐次远去，我寻一方避世之所，只为与自己独处。当外界纷扰渐远，这款陶瓷茶具，便成了我独享时光的静谧港湾，一个只属于我的精神避难所。\n\n它的精美设计，不是为了张扬，而是内敛地安抚着每一寸视线。温润的陶瓷手感，仿佛沉淀了千年岁月，每一次轻柔的触碰，都带着古朴的宁静与温度。泡一壶清茶，看茶叶在水中缓慢舒展，热气氤氲，模糊了窗外的一切尘嚣。这不是简单的饮茶，而是一场专属我的仪式感，是对慢生活最深沉的诠释。打开🎁精致的礼盒包装时，那份拆解惊喜便已将我带入另一个世界，连环保材质的选择，都透着一份对自然的敬意与体悟。\n\n这套茶具，不为迎合任何人的目光，只为安放我那颗渴望清净的灵魂。在茶香缭绕中，我思考，我沉淀，我与那个最真实的自己坦诚对话。它陪伴我度过无数个安静的午后，让孤独不再是空虚的代名词，而是丰盈自我的深刻体验。它是我在人群中抽离的理由，是我为自己打造的精神港湾。这份由茶具所赋予的独处，成为了我对抗世界喧嚣的武器，也成为了我内心最深处的慰藉与力量。\n\n如果你也渴望一份不被打扰的清寂，渴望在日常中寻觅一个避世之所，那么它，或许正是你一直在寻找的答案。让茶韵伴你，",
+      "tag": "HOT",
+      "target_audience": "茶文化爱好者"
+    },
+    {
+      "id": "C18",
+      "product_id": "P1",
+      "style_id": "S4",
+      "content": "城市里的喧嚣是不是常常让你感到心力交瘁？你有没有想过，能随时随地拥有一片只属于自己的宁静小天地？我的答案，是索尼WH-1000XM5降噪耳机。\n\n每天通勤的地铁上，人潮拥挤，嘈杂声此起彼伏，而我，却能安然享受着自己的一方天地。戴上它的那一刻，仿佛有一道无形的屏障轻轻落下，瞬间隔绝了所有噪音。那些烦躁的引擎声、人声鼎沸的交谈，都像被施了魔法般消失不见。只剩下纯粹的音乐，或是安静的留白，让思绪自由飘散，找到久违的平静。\n\n它的主动降噪功能真的太强大了，连微小的环境音都能被温柔地过滤掉。而且佩戴感简直是极致的舒适，轻盈得好像戴了片云朵，耳朵完全没有压迫感，即使是长途飞行或午后小憩，也能安心佩戴。超长续航也让人毫无后顾之忧，从早到晚都能有这份静谧相伴。\n\n无论是想沉浸在喜欢的音乐世界里，还是需要一个绝对安静的环境来专注工作、阅读，甚至是冥想，它都能完美满足。它不仅仅是一副耳机，更是我忙碌生活中的一份慰藉，",
+      "tag": "HOT",
+      "target_audience": "音乐爱好者"
+    },
+    {
+      "id": "C19",
+      "product_id": "P4",
+      "style_id": "S3",
+      "content": "深夜，万籁俱寂，只有屏幕微光映照着疲惫的脸。刷着那些高清到失真的完美瞬间，心头却涌上一股难以言喻的空虚。总觉得，现代世界的极致清晰，反而模糊了真实的情绪，是不是，我们都太习惯于追求无暇，却忘了残缺的美好？就在这样一个人静下来的夜晚，我遇见了它——这台复古CCD相机。\n\n它小巧轻便得能随意塞进包里，复古的外壳，握在手里仿佛就能触碰到旧时光的纹理。它不是为了挑战像素极限，而是想温柔地帮你留住那些转瞬即逝的“不完美”。按下快门，它捕捉到的不是冰冷的数字，而是带着温度的颗粒感，那种独特的低像素美学，让人瞬间掉进回忆的漩涡。\n\n深夜的咖啡馆，氤氲的水汽，窗外细密的雨丝，都被它赋予了一种独特的怀旧滤镜。低像素的照片，没有锐利的棱角，只有朦胧的氛围感，就像记忆深处那些被温柔模糊的旧梦。那些不经意间记录下的，或是深夜独酌的咖啡杯，或是路灯下被拉长的孤单身影，都因为这份独特的颗粒感，变得格外有故事，带着一丝感伤却又温柔的底色。\n\n它让我重新爱上那些模糊的、有点失焦的瞬间，因为它们更真实，更贴近心底的脆弱与感伤。每一张照片都像一封写给过去的信，又像一次与内心深处的对话，它不刻意，不喧哗，只是静静地，替你收藏起那些只有你自己懂得的，深夜情绪。带着它穿梭于城市边缘的寂静街道，那些被遗忘的角落，被它镀上了一层温柔的旧电影光晕。或许，我们需要的从来不是极致的清晰，而是能触动灵魂深处，那份带着温度的，独一无二的怀旧感。",
+      "tag": "HOT",
+      "target_audience": "摄影爱好者"
+    },
+    {
+      "id": "C20",
+      "product_id": "P5",
+      "style_id": "S4",
+      "content": "都市快节奏生活，你是不是也渴望片刻的宁静和自我对话？最近，我的治愈秘密武器就是这款手冲咖啡套装啦！它不仅是一套工具，更是一张通往慢生活的船票，带你体验独属的治愈时光。\n\n从研磨咖啡豆的沙沙声开始，到热水缓缓注入咖啡粉，看咖啡液如琥珀般一滴一滴坠落，整个过程都充满了庄重而美好的仪式感。精选的精品咖啡豆，打开瞬间就能闻到馥郁的香气，那是大自然最温柔的馈赠。一口下去，醇厚回甘，仿佛把一整个午后温暖的阳光都喝进了肚子里，瞬间洗去所有疲惫。\n\n即使你是咖啡小白也不用担心，专业的器具设计精巧，搭配详尽贴心的教程指导，让你轻松上手，享受成为自己专属咖啡师的乐趣。这份精致的包装，无论是送给自己一份心意，还是作为礼物馈赠好友，都显得格外有品味。独处的午后，一杯手冲，一本书，一段舒缓的音乐，所有的烦恼都被这温暖的香气温柔包裹，瞬间被治愈。它不只是一杯咖啡，更是一种慢下来、爱自己的生活态度。\n\n让手冲咖啡成为你生活中的小确幸，给自己一个安心舒适的治愈时刻吧。",
+      "tag": "HOT",
+      "target_audience": "咖啡爱好者"
+    }
+  ],
+  "relationships": [
+    {
+      "source": "P1",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "主动降噪",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P1",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "长续航",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P1",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "舒适佩戴",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P1",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "便携",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P1",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "静音设计",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P2",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "持久留香",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P2",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "天然蜡质",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P2",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "精致包装",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P2",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "环保材质",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P3",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "100%真丝",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P3",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "完全遮光",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P3",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "舒适贴合",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P3",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "便携",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P4",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "复古外观",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P4",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "颗粒感照片",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P4",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "轻便便携",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P4",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "便携",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P5",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "精品咖啡豆",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P5",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "专业器具",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P5",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "教程指导",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P5",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "精致包装",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P6",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "静音设计",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P6",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "LED夜灯",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P6",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "定时功能",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P7",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "护眼屏幕",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P7",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "长续航",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P7",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "海量图书",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P7",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "便携",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P8",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "防滑设计",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P8",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "环保材质",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P8",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "便携",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P8",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "舒适贴合",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P9",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "360度环绕",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P9",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "长续航",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P9",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "防水设计",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P10",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "精美设计",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P10",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "陶瓷材质",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P10",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "礼盒包装",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P10",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "精致包装",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P10",
+      "source_type": "Product",
+      "relationship": "HAS_FEATURE",
+      "target": "环保材质",
+      "target_type": "Feature"
+    },
+    {
+      "source": "P2",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C1",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P2",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C2",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P2",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C3",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P1",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C4",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P1",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C5",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P4",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C6",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P5",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C7",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P6",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C8",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P6",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C9",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P7",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C10",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P7",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C11",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P8",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C12",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P8",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C13",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P9",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C14",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P9",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C15",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P10",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C16",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P10",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C17",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P1",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C18",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P4",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C19",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "P5",
+      "source_type": "Product",
+      "relationship": "HAS_COPY",
+      "target": "C20",
+      "target_type": "Copywriting"
+    },
+    {
+      "source": "C1",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S1",
+      "target_type": "Style"
+    },
+    {
+      "source": "C2",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S2",
+      "target_type": "Style"
+    },
+    {
+      "source": "C3",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S3",
+      "target_type": "Style"
+    },
+    {
+      "source": "C4",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S1",
+      "target_type": "Style"
+    },
+    {
+      "source": "C5",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S2",
+      "target_type": "Style"
+    },
+    {
+      "source": "C6",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S2",
+      "target_type": "Style"
+    },
+    {
+      "source": "C7",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S1",
+      "target_type": "Style"
+    },
+    {
+      "source": "C8",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S4",
+      "target_type": "Style"
+    },
+    {
+      "source": "C9",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S1",
+      "target_type": "Style"
+    },
+    {
+      "source": "C10",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S4",
+      "target_type": "Style"
+    },
+    {
+      "source": "C11",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S1",
+      "target_type": "Style"
+    },
+    {
+      "source": "C12",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S4",
+      "target_type": "Style"
+    },
+    {
+      "source": "C13",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S1",
+      "target_type": "Style"
+    },
+    {
+      "source": "C14",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S2",
+      "target_type": "Style"
+    },
+    {
+      "source": "C15",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S3",
+      "target_type": "Style"
+    },
+    {
+      "source": "C16",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S5",
+      "target_type": "Style"
+    },
+    {
+      "source": "C17",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S1",
+      "target_type": "Style"
+    },
+    {
+      "source": "C18",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S4",
+      "target_type": "Style"
+    },
+    {
+      "source": "C19",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S3",
+      "target_type": "Style"
+    },
+    {
+      "source": "C20",
+      "source_type": "Copywriting",
+      "relationship": "HAS_STYLE",
+      "target": "S4",
+      "target_type": "Style"
+    },
+    {
+      "source": "C1",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "都市独居青年",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C2",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "年轻女性",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C3",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "深夜思考者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C4",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "通勤族",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C5",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "数码爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C6",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "摄影爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C7",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "咖啡爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C8",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "都市白领",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C9",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "都市独居青年",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C10",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "阅读爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C11",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "通勤族",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C12",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "瑜伽爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C13",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "瑜伽爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C14",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "音乐爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C15",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "深夜思考者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C16",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "茶文化爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C17",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "茶文化爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C18",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "音乐爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C19",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "摄影爱好者",
+      "target_type": "Persona"
+    },
+    {
+      "source": "C20",
+      "source_type": "Copywriting",
+      "relationship": "TARGET_AUDIENCE",
+      "target": "咖啡爱好者",
+      "target_type": "Persona"
+    }
+  ]
+}

rag_engine.py

@@ -0,0 +1,446 @@
+"""
+RAG引擎：实现传统RAG和GraphRAG的检索逻辑
+"""
+from typing import List, Dict, Tuple
+# 优先使用轻量级版本（避免超过 Vercel 250MB 限制）
+try:
+    from database_setup_lite import SimpleGraphDB, VectorDB
+except ImportError:
+    from database_setup import SimpleGraphDB, VectorDB
+import json
+import requests
+
+# LLM配置（从环境变量读取，确保安全）
+import os
+LLM_API_BASE = os.getenv("LLM_API_BASE", "https://api.ai-gaochao.cn/v1")
+LLM_API_KEY = os.getenv("LLM_API_KEY", "")
+LLM_MODEL = os.getenv("LLM_MODEL", "gemini-2.5-flash")
+
+if not LLM_API_KEY:
+    raise ValueError("LLM_API_KEY 环境变量未设置！请在 .env 文件中设置 LLM_API_KEY")
+
+class TraditionalRAG:
+    """传统语义RAG"""
+    def __init__(self, vector_db: VectorDB, graph_db: SimpleGraphDB = None):
+        self.vector_db = vector_db
+        self.graph_db = graph_db  # 用于限制搜索范围
+    
+    def retrieve(self, query: str, product_name: str = None, style_name: str = None, n_results: int = 5) -> Dict:
+        """语义检索（传统RAG：直接向量搜索，不利用图结构，返回片段句子）"""
+        # 传统RAG的特点：直接进行语义相似度搜索，不利用图结构
+        # 使用相同的文案数据库，但只返回相似的片段句子（而不是完整文案）
+        
+        # 直接进行向量搜索（传统RAG的特点）
+        # 传统RAG限制结果数量，只返回最相关的2-3个结果
+        limited_results = min(3, n_results)  # 最多返回3个结果
+        all_results = self.vector_db.search(query, n_results=limited_results * 2)  # 多搜索一些，用于提取片段
+        
+        # 从完整文案中提取与查询最相关的片段句子
+        processed_results = []
+        query_keywords = set(query.lower().split())
+        
+        for result in all_results[:limited_results * 2]:
+            full_content = result.get("content", "")
+            if not full_content:
+                continue
+            
+            # 将文案按句子分割（中文句号、英文句号、感叹号、问号）
+            import re
+            sentences = re.split(r'[。！？.!?]', full_content)
+            sentences = [s.strip() for s in sentences if s.strip()]
+            
+            # 找到与查询最相关的句子片段
+            best_sentences = []
+            for sentence in sentences:
+                # 计算句子与查询的相关度（简单关键词匹配）
+                sentence_lower = sentence.lower()
+                keyword_matches = sum(1 for keyword in query_keywords if keyword in sentence_lower)
+                if keyword_matches > 0:
+                    best_sentences.append((sentence, keyword_matches))
+            
+            # 按相关度排序，取前2-3个最相关的句子
+            best_sentences.sort(key=lambda x: x[1], reverse=True)
+            selected_sentences = [s[0] for s in best_sentences[:3]]
+            
+            # 如果没有找到相关句子，取前3个句子作为片段
+            if not selected_sentences and sentences:
+                selected_sentences = sentences[:3]
+            
+            # 组合成片段（最多150字，确保有足够内容）
+            snippet = "。".join(selected_sentences)
+            if not snippet and sentences:
+                # 如果还是空的，至少取前3个句子
+                snippet = "。".join(sentences[:3])
+            if len(snippet) > 150:
+                snippet = snippet[:150] + "..."
+            elif len(snippet) < 30 and len(sentences) > 0:
+                # 如果片段太短，至少取前2-3个句子
+                snippet = "。".join(sentences[:min(3, len(sentences))])
+                if len(snippet) > 150:
+                    snippet = snippet[:150] + "..."
+            
+            if snippet:
+                processed_results.append({
+                    "content": snippet,  # 返回片段而不是完整文案
+                    "full_content": full_content,  # 保留完整内容用于显示
+                    "metadata": result.get("metadata", {}),
+                    "distance": result.get("distance", 0),
+                    "is_snippet": True  # 标记这是片段
+                })
+            
+            if len(processed_results) >= limited_results:
+                break
+        
+        # 如果结果太少，至少返回1-2个语义相似的结果
+        if len(processed_results) < 1:
+            # 如果提取片段失败，至少返回一些结果
+            for result in all_results[:max(1, limited_results)]:
+                content = result.get("content", "")
+                if content:
+                    # 简单截取前150字作为片段
+                    snippet = content[:150] + "..." if len(content) > 150 else content
+                    processed_results.append({
+                        "content": snippet,
+                        "full_content": content,
+                        "metadata": result.get("metadata", {}),
+                        "distance": result.get("distance", 0),
+                        "is_snippet": True
+                    })
+                if len(processed_results) >= limited_results:
+                    break
+        
+        return {
+            "method": "语义检索",
+            "query": query,
+            "product": product_name,
+            "style": style_name,
+            "results": processed_results[:limited_results],
+            "retrieval_path": [
+                "向量相似度搜索（传统RAG：不利用图结构）",
+                f"找到 {len(processed_results)} 个语义相似的片段",
+                "⚠️  局限性：只返回片段句子，没有图结构，无法找到跨品类的风格相关文案"
+            ],
+            "explanation": "传统RAG直接通过语义相似度搜索相关文案，使用相同的文案数据库，但只返回与查询最相关的片段句子（而不是完整文案）。没有图结构，无法找到跨品类的风格相关文案。"
+        }
+
+class GraphRAG:
+    """图增强RAG"""
+    def __init__(self, graph_db: SimpleGraphDB, vector_db: VectorDB):
+        self.graph_db = graph_db
+        self.vector_db = vector_db
+    
+    def retrieve(self, query: str, product_name: str = None, style_name: str = None, n_results: int = 5) -> Dict:
+        """图增强检索"""
+        retrieval_path = []
+        retrieved_docs = []
+        
+        # 步骤1: 尝试找到风格节点
+        style_node = None
+        if style_name:
+            style_node = self.graph_db.find_node_by_property("Style", "name", style_name)
+            if style_node:
+                retrieval_path.append(f"定位风格节点: {style_node['properties']['name']}")
+        
+        # 步骤2: 通过风格节点找到相关文案（跨品类）
+        if style_node:
+            # 反向查找：找到连接到风格的文案节点
+            for edge in self.graph_db.edges:
+                if edge["target"] == style_node["id"] and edge["relationship"] == "HAS_STYLE":
+                    copy_node = self.graph_db.nodes.get(edge["source"])
+                    if copy_node and copy_node["type"] == "Copywriting":
+                        content = copy_node["properties"]["content"]
+                        # 获取该文案关联的产品（HAS_COPY关系：Product -> Copywriting）
+                        product_id = None
+                        for e in self.graph_db.edges:
+                            if e["target"] == edge["source"] and e["relationship"] == "HAS_COPY":
+                                product_id = e["source"]
+                                break
+                        
+                        product_info = self.graph_db.nodes.get(product_id, {}).get("properties", {})
+                        retrieved_docs.append({
+                            "content": content,
+                            "source": "图遍历",
+                            "product": product_info.get("name", "未知"),
+                            "style": style_name,
+                            "tag": copy_node["properties"].get("tag", ""),
+                            "retrieval_reason": f"通过风格节点'{style_name}'找到的跨品类文案（来自产品：{product_info.get('name', '未知')}）"
+                        })
+            
+            if retrieved_docs:
+                retrieval_path.append(f"通过风格节点遍历找到 {len(retrieved_docs)} 个相关文案")
+            else:
+                retrieval_path.append("未找到该风格的相关文案")
+        
+        # 步骤3: 如果指定了产品，查找产品特征
+        product_features = []
+        if product_name:
+            product_node = self.graph_db.find_node_by_property("Product", "name", product_name)
+            if product_node:
+                retrieval_path.append(f"定位产品节点: {product_name}")
+                features = product_node["properties"].get("features", [])
+                keywords = product_node["properties"].get("keywords", [])
+                product_features = features + keywords
+                retrieval_path.append(f"提取产品特征: {', '.join(product_features[:5])}")
+        
+        # 步骤4: 如果图检索结果不足，用向量检索补充
+        if len(retrieved_docs) < n_results:
+            vector_results = self.vector_db.search(query, n_results=n_results - len(retrieved_docs))
+            for result in vector_results:
+                # 避免重复
+                if not any(doc["content"] == result["content"] for doc in retrieved_docs):
+                    retrieved_docs.append({
+                        "content": result["content"],
+                        "source": "向量检索补充",
+                        "product": result["metadata"].get("product_id", "未知"),
+                        "style": result["metadata"].get("style_id", "未知"),
+                        "tag": result["metadata"].get("tag", ""),
+                        "retrieval_reason": "语义���似度补充检索"
+                    })
+            if vector_results:
+                retrieval_path.append(f"向量检索补充 {len(vector_results)} 个结果")
+        
+        return {
+            "method": "图增强检索",
+            "query": query,
+            "product": product_name,
+            "style": style_name,
+            "product_features": product_features,
+            "results": retrieved_docs[:n_results],
+            "retrieval_path": retrieval_path,
+            "explanation": "通过图结构找到跨品类的风格相关文案，即使产品不同，但风格相通，可以借鉴文案模板。"
+        }
+
+class RAGEngine:
+    """RAG引擎主类"""
+    def __init__(self, graph_db: SimpleGraphDB, vector_db: VectorDB):
+        self.graph_db = graph_db
+        self.traditional_rag = TraditionalRAG(vector_db, graph_db)
+        self.graph_rag = GraphRAG(graph_db, vector_db)
+    
+    def compare_retrieval(self, query: str, product_name: str = None, style_name: str = None) -> Dict:
+        """对比传统RAG和GraphRAG的检索结果"""
+        traditional_result = self.traditional_rag.retrieve(query, product_name, style_name)
+        graph_result = self.graph_rag.retrieve(query, product_name, style_name)
+        
+        return {
+            "traditional_rag": traditional_result,
+            "graph_rag": graph_result,
+            "comparison": {
+                "traditional_count": len(traditional_result["results"]),
+                "graph_count": len(graph_result["results"]),
+                "graph_cross_category": len([r for r in graph_result["results"] if r.get("source") == "图遍历"])
+            }
+        }
+    
+    def generate_copywriting(self, query: str, product_name: str, style_name: str, use_graph: bool = True) -> Dict:
+        """生成文案（使用LLM）"""
+        if use_graph:
+            retrieval_result = self.graph_rag.retrieve(query, product_name, style_name)
+        else:
+            retrieval_result = self.traditional_rag.retrieve(query, product_name, style_name)
+        
+        # 获取检索到的参考文案
+        retrieved_texts = [r["content"] for r in retrieval_result["results"][:5]]  # 取前5个作为参考
+        
+        # 统计信息
+        cross_category_count = len([r for r in retrieval_result["results"] if r.get("source") == "图遍历"]) if use_graph else 0
+        
+        # 获取产品特征（用于GraphRAG）
+        product_features = []
+        if use_graph and retrieval_result.get("product_features"):
+            product_features = retrieval_result["product_features"]
+        
+        # 调用LLM生成文案
+        try:
+            llm_generated = self._call_llm_generate(
+                product_name=product_name,
+                style_name=style_name,
+                reference_texts=retrieved_texts,
+                product_features=product_features,
+                use_graph=use_graph,
+                cross_category_count=cross_category_count
+            )
+        except Exception as e:
+            print(f"LLM生成失败: {e}")
+            # 如果LLM失败，使用模板生成
+            llm_generated = self._generate_template(retrieved_texts, product_name, style_name)
+        
+        # 组装最终输出
+        if use_graph and product_features:
+            features = ", ".join(product_features[:3])
+            reference_sources = ', '.join([r.get('product', '未知') for r in retrieval_result["results"][:3]])
+            generated_text = f"""基于图增强检索生成的文案：
+
+✨ 检索策略：通过图结构找到跨品类的风格相关文案
+📊 检索结果：找到 {len(retrieved_texts)} 个相关文案，其中 {cross_category_count} 个来自跨品类（通过风格节点关联）
+🎯 产品特征：{features}
+📝 参考文案来源：{reference_sources}
+
+【{style_name}风格】{product_name}文案：
+
+{llm_generated}
+
+💡 说明：GraphRAG 通过风格节点找到了跨品类的参考文案（如香薰蜡烛的清冷避世风文案），即使产品不同，但风格相通，可以借鉴文案模板。"""
+        else:
+            generated_text = f"""基于传统语义检索生成的文案：
+
+🔍 检索策略：直接通过语义相似度搜索
+📊 检索结果：找到 {len(retrieved_texts)} 个语义相似的文案
+⚠️  局限性：如果数据库中没有相似内容，可能返回不相关的结果
+
+【{style_name}风格】{product_name}文案：
+
+{llm_generated}
+
+💡 说明：传统 RAG 只能找到语义相似的文案，如果数据库中没有该产品的该风格文案，可能无法生成合适的文案。"""
+        
+        return {
+            "generated_text": generated_text,
+            "retrieval_result": retrieval_result,
+            "method": "GraphRAG" if use_graph else "Traditional RAG"
+        }
+    
+    def _call_llm_generate(self, product_name: str, style_name: str, reference_texts: List[str], 
+                           product_features: List[str] = None, use_graph: bool = True, 
+                           cross_category_count: int = 0) -> str:
+        """调用LLM生成文案"""
+        headers = {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {LLM_API_KEY}"
+        }
+        url = f"{LLM_API_BASE}/chat/completions"
+        
+        # 构建参考文案说明
+        reference_context = ""
+        if reference_texts:
+            reference_context = "\n\n参考文案（用于学习风格和句式）：\n"
+            for i, text in enumerate(reference_texts[:3], 1):
+                reference_context += f"{i}. {text}\n"
+        else:
+            reference_context = "\n\n⚠️ 注意：没有找到相关参考文案，请根据产品特征和风格要求创作。"
+        
+        # 构建产品特征说明
+        features_context = ""
+        if product_features:
+            features_context = f"\n产品特征：{', '.join(product_features[:5])}"
+        
+        # 构建prompt
+        if use_graph and cross_category_count > 0:
+            prompt = f"""你是一名擅长小红书文案写作的创意编辑。请根据以下信息，生成一篇适合在小红书发布的文案（200-300字，要求内容丰富、有细节感）。
+
+产品名称：{product_name}
+目标风格：{style_name}
+{features_context}
+
+{reference_context}
+
+重要提示：
+1. 这些参考文案来自其他产品（跨品类），但风格相同，请学习它们的句式、语气和情感表达方式
+2. 将参考文案的风格和句式应用到目标产品上
+3. 文案要有细节感、人情味，符合小红书用户的阅读习惯
+4. 保持{style_name}的风格特征
+5. 文案长度要求200-300字，要有丰富的内容和细节描述，可以包含使用场景、情感体验、产品特色等多个方面
+6. 请确保文案完整，不要被截断，以完整的句子结尾
+
+请直接输出文案内容，不要包含"好的"、"没问题"等前缀，也不要使用markdown格式。只输出文案正文，确保内容完整。"""
+        else:
+            prompt = f"""你是一名擅长小红书文案写作的创意编辑。请根据以下信息，生成一篇适合在小红书发布的文案（200-300字，要求内容丰富、有细节感）。
+
+产品名称：{product_name}
+目标风格：{style_name}
+{features_context}
+
+{reference_context}
+
+重要提示：
+1. 参考文案可能有限或不够相关，请根据产品特征和风格要求创作
+2. 文案要有细节感、人情味，符合小红书用户的阅读习惯
+3. 保持{style_name}的风格特征
+4. 文案长度要求200-300字，要有丰富的内容和细节描述，可以包含使用场景、情感体验、产品特色等多个方面
+5. 请确保文案完整，不要被截断，以完整的句子结尾
+
+请直接输出文案内容，不要包含"好的"、"没问题"等前缀，也不要使用markdown格式。只输出文案正文，确保内容完整。"""
+        
+        body = {
+            "model": LLM_MODEL,
+            "messages": [
+                {
+                    "role": "system",
+                    "content": "你是一名擅长文案写作的创意编辑，擅长创作小红书风格的文案。"
+                },
+                {
+                    "role": "user",
+                    "content": prompt
+                }
+            ],
+            "max_tokens": 4000,  # 增加token限制以支持更长的文案（200-300字约需要800-1200 tokens，设置4000确保完整输出）
+            "temperature": 0.9
+        }
+        
+        resp = requests.post(url, headers=headers, json=body, timeout=60)
+        resp.raise_for_status()
+        data = resp.json()
+        generated = data["choices"][0]["message"]["content"].strip()
+        
+        # 清理生成的内容
+        # 移除常见的前缀（只移除开头的前缀，不要截断内容）
+        prefixes_to_remove = [
+            "好的，没问题！",
+            "好的，",
+            "没问题！",
+            "好的！",
+        ]
+        for prefix in prefixes_to_remove:
+            if generated.startswith(prefix):
+                generated = generated[len(prefix):].strip()
+        
+        # 移除markdown格式符号（但保留内容）
+        generated = generated.replace("**", "").replace("*", "").strip()
+        
+        return generated
+    
+    def _generate_template(self, reference_texts: List[str], product_name: str, style_name: str) -> str:
+        """生成文案模板（简化版，实际应调用LLM）"""
+        # 如果有参考文案，提取关键句式
+        key_phrases = []
+        if reference_texts:
+            for text in reference_texts[:2]:  # 只取前2个参考
+                # 提取关键句式（简单提取）
+                if "避难所" in text:
+                    key_phrases.append("避难所")
+                if "安静" in text:
+                    key_phrases.append("安静")
+                if "唯一" in text:
+                    key_phrases.append("唯一")
+                if "���绝子" in text:
+                    key_phrases.append("绝绝子")
+        
+        # 根据风格和产品生成
+        if "清冷避世风" in style_name or "深夜emo" in style_name.lower():
+            if "眼罩" in product_name:
+                if key_phrases:
+                    # GraphRAG：使用参考文案的句式
+                    return f"戴上眼罩的这片刻漆黑，是我在繁杂城市里唯一的{'避难所' if '避难所' in key_phrases else '避风港'}。物理意义上的关灯，也是心理上的断联。世界终于{'安静了' if '安静' in key_phrases else '静下来了'}，今晚只属于我自己。"
+                else:
+                    # 传统RAG：没有参考，使用通用模板
+                    return f"这个{product_name}真的很不错，遮光效果好，推荐给大家使用。"
+            elif "CCD" in product_name or "相机" in product_name:
+                return "深夜拿起它，在颗粒感的画面里，所有的情绪都有了出口。低像素不是缺陷，是另一种真实。"
+            else:
+                if key_phrases:
+                    return f"每一个与{product_name}的瞬间，都是我与世界的{'唯一连接' if '唯一' in key_phrases else '连接'}。"
+                else:
+                    return f"这个{product_name}真的很不错，推荐给大家。"
+        elif "疯狂种草" in style_name:
+            if key_phrases and "绝绝子" in key_phrases:
+                # GraphRAG：使用参考文案的语气
+                return f"家人们谁懂啊！这个{product_name}真的绝绝子，一秒沦陷！必须人手一个！"
+            else:
+                # 传统RAG：没有参考，使用通用语气
+                return f"这个{product_name}真的很不错，推荐给大家购买！"
+        else:
+            if key_phrases:
+                return f"这个{product_name}真的很不错，{'强烈推荐' if '绝绝子' in key_phrases else '推荐'}给大家！"
+            else:
+                return f"这个{product_name}真的很不错，推荐给大家！"
+

requirements.txt

@@ -0,0 +1,8 @@
+fastapi>=0.104.0
+uvicorn[standard]>=0.24.0
+pydantic>=2.0.0
+requests>=2.31.0
+python-multipart>=0.0.6
+chromadb>=0.4.22
+numpy>=1.24.0
+openai>=1.0.0