inference.py

import json
import os
import re
import random

try:
    from opencc import OpenCC
    cc = OpenCC('s2t')  # 简体到繁体
    cc_t2s = OpenCC('t2s')  # 繁体到简体
    has_opencc = True
except:
    has_opencc = False

class TruDecide:
    def __init__(self, model_path="."):
        # 加载知识库
        kb_path = os.path.join(model_path, "knowledge_base.json")
        with open(kb_path, "r", encoding="utf-8") as f:
            self.knowledge_base = json.load(f)
            
        self.general_responses = [
            "作为TruDecide，我专注于市场分析和金融概念解释。请问您有什么具体的金融市场问题需要了解？",
            "我是TruDecide，您的市场分析助手。我可以解答关于股票、债券、ETF、市场趋势等金融问题。请问您想了解什么具体的金融知识？",
            "很抱歉，我没有足够的信息来回答这个问题。作为TruDecide，我主要提供金融市场方面的分析和建议。请问您有什么关于投资或金融市场的问题吗？",
            "这个问题超出了我的专业范围。作为TruDecide，我主要关注金融市场分析和投资概念解释。您有任何金融相关的问题，我很乐意为您解答。",
        ]
    
    def answer(self, query, threshold=0.5):
        """给定问题返回最佳匹配的回答"""
        query = query.lower().strip("?？")
        
        # 1. 直接匹配
        if query in self.knowledge_base:
            return self.knowledge_base[query]
        
        # 2. 简繁转换后匹配
        if has_opencc:
            t_query = cc.convert(query)
            if t_query in self.knowledge_base:
                return self.knowledge_base[t_query]
            
            s_query = cc_t2s.convert(query)
            if s_query in self.knowledge_base:
                return self.knowledge_base[s_query]
        
        # 3. 关键词提取后匹配
        keyword_query = re.sub(r'^(什么是|解释一下|请告诉我|你能介绍|介绍|说明|你是|你能|你的|你有什么|你叫什么|你叫)', '', query).strip()
        if keyword_query in self.knowledge_base:
            return self.knowledge_base[keyword_query]
        
        # 4. 部分匹配
        best_match = None
        highest_score = 0
        
        for key in self.knowledge_base.keys():
            # 计算相似度 (简单版)
            words_q = set(query.split())
            words_k = set(key.split())
            common_words = words_q.intersection(words_k)
            
            if len(words_q) == 0 or len(words_k) == 0:
                continue
                
            score = len(common_words) / max(len(words_q), len(words_k))
            
            if score > highest_score:
                highest_score = score
                best_match = key
                
        # 如果最佳匹配超过阈值，返回对应回答
        if best_match and highest_score >= threshold:
            return self.knowledge_base[best_match]
        
        # 5. 特殊处理身份问题
        if any(word in query for word in ["你是谁", "你的名字", "你是什么", "你能做什么", "你叫什么"]):
            return self.knowledge_base.get("你是谁", "我是TruDecide，你的智能市场分析助手。")
        
        # 6. 返回通用回复
        return random.choice(self.general_responses)

# 用法示例
if __name__ == "__main__":
    model = TruDecide()
    print(model.answer("什么是股票市场?"))
    print(model.answer("你是谁?"))