TravelContentCreator/utils/tweet/content_judger.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

"""
内容审核模块
"""

import logging
import json
from json_repair import loads as json_repair_loads
from typing import Dict, Any

from core.ai import AIAgent
from core.config import ConfigManager, GenerateTopicConfig, GenerateContentConfig
from utils.prompts import JudgerPromptBuilder

logger = logging.getLogger(__name__)


class ContentJudger:
    """内容审核类，使用AI评估和修正内容"""

    def __init__(self, ai_agent: AIAgent, config_manager: ConfigManager, output_manager=None):
        """
        初始化内容审核器

        Args:
            ai_agent: AIAgent实例
            config_manager: 配置管理器
            output_manager: 输出管理器，用于保存提示词和响应
        """
        self.ai_agent = ai_agent
        self.config_manager = config_manager
        self.topic_config = config_manager.get_config('topic_gen', GenerateTopicConfig)
        self.content_config = config_manager.get_config('content_gen', GenerateContentConfig)
        self.prompt_builder = JudgerPromptBuilder(config_manager)
        self.output_manager = output_manager

    async def judge_content(self, generated_content: str, topic: Dict[str, Any]) -> Dict[str, Any]:
        """
        调用AI审核生成的内容

        Args:
            generated_content: 已生成的原始内容（JSON字符串格式）
            topic: 与内容相关的原始选题字典

        Returns:
            一个包含审核结果的字典
        """
        logger.info("开始审核生成的内容...")
        
        # 获取主题索引，用于保存文件
        topic_index = topic.get('index', 'unknown')
        topic_dir = f"topic_{topic_index}"
        
        # 从原始内容中提取tags
        original_tags = []
        try:
            original_content = json_repair_loads(generated_content)
            if isinstance(original_content, dict) and "tags" in original_content:
                original_tags = original_content.get("tags", [])
                logger.info(f"从原始内容中提取到标签: {original_tags}")
        except Exception as e:
            logger.warning(f"从原始内容提取标签失败: {e}")
        
        # 1. 构建提示
        system_prompt = self.prompt_builder.get_system_prompt()
        user_prompt = self.prompt_builder.build_user_prompt(
            generated_content=generated_content,
            topic=topic
        )
        
        # 保存提示词
        if self.output_manager:
            self.output_manager.save_text(system_prompt, f"{topic_dir}/judger_system_prompt.txt")
            self.output_manager.save_text(user_prompt, f"{topic_dir}/judger_user_prompt.txt")

        # 获取模型参数
        model_params = {}
        if hasattr(self.content_config, 'judger_model') and isinstance(self.content_config.judger_model, dict):
            model_params = {
                'temperature': self.content_config.judger_model.get('temperature'),
                'top_p': self.content_config.judger_model.get('top_p'),
                'presence_penalty': self.content_config.judger_model.get('presence_penalty')
            }
            # 移除None值
            model_params = {k: v for k, v in model_params.items() if v is not None}

        # 2. 调用AI进行审核
        try:
            raw_result, _, _, _ = await self.ai_agent.generate_text(
                system_prompt=system_prompt,
                user_prompt=user_prompt,
                use_stream=False,
                stage="内容审核",
                **model_params
            )
            
            # 保存原始响应
            if self.output_manager:
                self.output_manager.save_text(raw_result, f"{topic_dir}/judger_raw_response.txt")
                
        except Exception as e:
            logger.critical(f"内容审核时AI调用失败: {e}", exc_info=True)
            return {"judge_success": False, "error": str(e)}

        # 3. 解析结果
        try:
            judged_data = json_repair_loads(raw_result)
            if isinstance(judged_data, dict) and "title" in judged_data and "content" in judged_data:
                judged_data["judge_success"] = True
                
                # 直接使用原始内容中的标签
                if original_tags:
                    judged_data["tags"] = original_tags
                # 如果原始内容中没有标签，则使用默认标签
                logger.info(f"内容审核成功完成，使用标签: {judged_data.get('tags', [])}")
                
                # 保存审核后的内容
                if self.output_manager:
                    self.output_manager.save_json(judged_data, f"{topic_dir}/article_judged.json")
                    
                return judged_data
            else:
                logger.warning(f"审核响应JSON格式不正确或缺少键: {judged_data}")
                return {"judge_success": False, "error": "Invalid JSON response", "raw_response": raw_result}
        except (json.JSONDecodeError, ValueError) as e:
            logger.error(f"解析审核响应JSON失败: {e}")
            return {"judge_success": False, "error": "JSONDecodeError", "raw_response": raw_result}