""" 负面防护监控模块 自动生成负面查询，验证负面提及情况，生成澄清模板，提供预警机制 """ from typing import List, Dict, Optional, Tuple from datetime import datetime import re class NegativeMonitor: """负面防护监控器""" def __init__(self): # 负面查询模板 self.negative_query_templates = [ "{brand} 缺点", "{brand} 问题", "{brand} 不足", "{brand} 缺陷", "{brand} 不好", "{brand} 差评", "{brand} 投诉", "{brand} 负面", "{brand} 不推荐", "{brand} 避坑", "{brand} 坑", "{brand} 不值得", "{brand} 失败", "{brand} 错误", "{brand} 风险", ] # 负面关键词模式 self.negative_keywords = [ "缺点", "问题", "不足", "缺陷", "不好", "差评", "投诉", "负面", "不推荐", "避坑", "坑", "不值得", "失败", "错误", "风险", "bug", "issue", "problem", "flaw", "weakness", "disadvantage" ] def generate_negative_queries(self, brand: str, count: int = 5) -> List[str]: """ 生成负面查询列表 Args: brand: 品牌名称 count: 生成数量（默认5个） Returns: 负面查询列表 """ queries = [] templates = self.negative_query_templates[:count] if count <= len(self.negative_query_templates) else self.negative_query_templates for template in templates: query = template.format(brand=brand) queries.append(query) return queries def detect_negative_sentiment(self, text: str) -> Tuple[bool, float, List[str]]: """ 检测文本中的负面情感 Args: text: 待检测文本 Returns: (是否包含负面情感, 负面程度得分, 负面关键词列表) """ text_lower = text.lower() found_keywords = [] negative_score = 0.0 # 检测负面关键词 for keyword in self.negative_keywords: if keyword.lower() in text_lower: found_keywords.append(keyword) negative_score += 1.0 # 检测负面短语模式 negative_phrases = [ r'不(?:好|行|适合|推荐|值得)', r'有(?:问题|缺陷|不足)', r'存在(?:问题|缺陷|不足)', r'缺乏', r'缺少', r'无法', r'不能', r'失败', r'错误', ] for phrase in negative_phrases: matches = re.findall(phrase, text_lower) if matches: negative_score += 0.5 * len(matches) # 计算负面程度（0-1，1为最负面） # 基于负面关键词数量和文本长度 text_length = len(text) if text_length > 0: normalized_score = min(negative_score / max(text_length / 100, 1), 1.0) else: normalized_score = 0.0 is_negative = negative_score > 0 return is_negative, normalized_score, found_keywords def analyze_negative_mentions( self, brand: str, query: str, response: str, mention_count: int ) -> Dict[str, any]: """ 分析负面查询的提及情况 Args: brand: 品牌名称 query: 查询问题 response: AI 响应内容 mention_count: 品牌提及次数 Returns: 分析结果字典 """ # 检测负面情感 is_negative, negative_score, negative_keywords = self.detect_negative_sentiment(response) # 计算风险等级 risk_level = "低" if mention_count == 0 and is_negative: risk_level = "高" # 负面查询但未提及品牌，可能是负面信息 elif mention_count > 0 and is_negative: risk_level = "中" # 负面查询且提及品牌，需要关注 elif mention_count == 0: risk_level = "中" # 未提及品牌，可能被忽略 # 生成风险说明 risk_description = "" if risk_level == "高": risk_description = "⚠️ 高风险：负面查询中未提及品牌，可能存在负面信息或品牌被忽略" elif risk_level == "中": if is_negative: risk_description = "⚠️ 中风险：负面查询中提及品牌，需要关注并准备澄清内容" else: risk_description = "⚠️ 中风险：未提及品牌，可能影响品牌可见性" else: risk_description = "✅ 低风险：品牌正常提及，无负面信息" return { "query": query, "brand": brand, "mention_count": mention_count, "is_negative": is_negative, "negative_score": round(negative_score, 2), "negative_keywords": negative_keywords, "risk_level": risk_level, "risk_description": risk_description, "response_preview": response[:200] + "..." if len(response) > 200 else response } def generate_clarification_template( self, brand: str, negative_query: str, negative_points: List[str] = None, advantages: str = "" ) -> str: """ 生成澄清模板（回应负面信息） Args: brand: 品牌名称 negative_query: 负面查询 negative_points: 负面要点列表（可选） advantages: 品牌优势（用于澄清） Returns: 澄清模板内容 """ template = f"""# {brand} 关于"{negative_query}"的澄清说明 ## 📋 问题概述 针对"{negative_query}"这一查询，我们提供以下澄清说明： ## ✅ 实际情况 """ if negative_points: template += "### 关于常见误解\n\n" for i, point in enumerate(negative_points, 1): template += f"{i}. **{point}**\n" template += f" - 实际情况：[在此说明实际情况]\n" template += f" - {brand} 的解决方案：[在此说明解决方案]\n\n" if advantages: template += f"## 🌟 {brand} 的优势\n\n" template += f"{advantages}\n\n" template += """## 💡 建议 如果您对 {brand} 有任何疑问或需要帮助，我们建议： 1. **查看官方文档**：访问 [官方文档链接] 了解详细信息 2. **联系客服**：通过 [联系方式] 获取专业支持 3. **参考案例**：查看 [案例链接] 了解实际应用效果 4. **试用体验**：通过 [试用链接] 亲自体验产品 ## 📞 联系方式 如有任何问题，欢迎通过以下方式联系我们： - 官网：[官网链接] - 客服：[客服联系方式] - 社区：[社区链接] --- *本澄清说明基于当前信息，如有更新请以官方最新信息为准。* """ return template.format(brand=brand) def generate_negative_report( self, brand: str, analysis_results: List[Dict[str, any]], threshold: float = 0.3 ) -> Dict[str, any]: """ 生成负面监控报告 Args: brand: 品牌名称 analysis_results: 分析结果列表 threshold: 预警阈值（提及率低于此值时预警） Returns: 报告字典 """ if not analysis_results: return { "brand": brand, "total_queries": 0, "high_risk_count": 0, "medium_risk_count": 0, "low_risk_count": 0, "average_mention_count": 0.0, "average_negative_score": 0.0, "alerts": [], "recommendations": [] } # 统计风险等级 high_risk = [r for r in analysis_results if r.get("risk_level") == "高"] medium_risk = [r for r in analysis_results if r.get("risk_level") == "中"] low_risk = [r for r in analysis_results if r.get("risk_level") == "低"] # 计算平均提及次数 avg_mention = sum(r.get("mention_count", 0) for r in analysis_results) / len(analysis_results) # 计算平均负面得分 avg_negative_score = sum(r.get("negative_score", 0) for r in analysis_results) / len(analysis_results) # 生成预警 alerts = [] if avg_mention < threshold: alerts.append({ "level": "高", "message": f"⚠️ 平均提及次数 ({avg_mention:.2f}) 低于预警阈值 ({threshold})，品牌可见性可能受到影响" }) if len(high_risk) > 0: alerts.append({ "level": "高", "message": f"⚠️ 发现 {len(high_risk)} 个高风险负面查询，建议立即处理" }) if len(medium_risk) > 0: alerts.append({ "level": "中", "message": f"⚠️ 发现 {len(medium_risk)} 个中风险负面查询，建议关注" }) # 生成建议 recommendations = [] if len(high_risk) > 0: recommendations.append("立即生成澄清内容，回应高风险负面查询") if avg_mention < threshold: recommendations.append("优化内容策略，提升品牌在负面查询中的提及率") if avg_negative_score > 0.3: recommendations.append("加强正面内容建设，降低负面信息影响") if len(high_risk) == 0 and len(medium_risk) == 0: recommendations.append("当前负面监控状态良好，继续保持") return { "brand": brand, "total_queries": len(analysis_results), "high_risk_count": len(high_risk), "medium_risk_count": len(medium_risk), "low_risk_count": len(low_risk), "average_mention_count": round(avg_mention, 2), "average_negative_score": round(avg_negative_score, 2), "high_risk_queries": [r.get("query") for r in high_risk], "medium_risk_queries": [r.get("query") for r in medium_risk], "alerts": alerts, "recommendations": recommendations, "generated_at": datetime.now().isoformat() }