""" 优化建议引擎 - 基于品牌数据差距提供可操作的优化建议 建议类型: - content_optimization: 内容优化建议（如何让AI更容易引用你的内容） - platform_targeting: 平台定向建议（哪些平台需要重点优化） - competitor_gap: 竞品差距建议（如何缩小与竞品的差距） - query_expansion: 查询词扩展建议（应该添加哪些查询词） - citation_improvement: 引用改善建议（如何增加被引用的概率） 生成逻辑: 1. 基于评分V2的五维度数据，找出最弱的维度 2. 基于竞品对比数据，找出差距最大的方面 3. 基于情感分析数据，找出负面情感的来源 4. 使用DeepSeek API生成个性化建议（ENABLE_LLM控制开关） 5. 无LLM时使用基于规则的模板建议 """ from __future__ import annotations import asyncio import json import logging import uuid from dataclasses import dataclass, field from typing import Any from app.config import settings from app.services.scoring.scoring_service import ScoringResultV2 from app.utils.json_extractor import extract_json logger = logging.getLogger(__name__) # ============================================================ # 建议数据结构 # ============================================================ @dataclass class SuggestionItem: """单条优化建议""" type: str # content_optimization/platform_targeting/competitor_gap/query_expansion/citation_improvement priority: str # high/medium/low title: str # 建议标题 description: str # 详细描述 action: str # 具体操作步骤 expected_impact: str # 预期效果 difficulty: str # easy/medium/hard @dataclass class BrandAnalysisContext: """品牌分析上下文 - 用于生成建议的输入数据""" brand_name: str overall_score: float # 五维度评分 mention_rate_score: float = 0.0 mention_rate_max: float = 25.0 mention_rate_percentage: float = 0.0 rank_score: float = 0.0 rank_max: float = 25.0 rank_percentage: float = 0.0 sentiment_score: float = 0.0 sentiment_max: float = 20.0 sentiment_percentage: float = 0.0 citation_score: float = 0.0 citation_max: float = 15.0 citation_percentage: float = 0.0 competitive_score: float = 0.0 competitive_max: float = 15.0 competitive_percentage: float = 0.0 # 竞品对比数据 competitor_data: dict[str, Any] = field(default_factory=dict) # 情感分析数据 sentiment_data: dict[str, int] = field(default_factory=dict) # 平台评分数据 platform_scores: dict[str, float] = field(default_factory=dict) # 查询词数据 total_queries: int = 0 mentioned_count: int = 0 # ============================================================ # 基于规则的建议生成器 # ============================================================ def _get_weakest_dimensions(ctx: BrandAnalysisContext) -> list[tuple[str, float, float]]: """ 找出最弱的维度，按得分率升序排列。 Returns: [(维度名, 得分率, 得分), ...] """ dimensions = [ ("提及率", ctx.mention_rate_percentage, ctx.mention_rate_score), ("推荐排名", ctx.rank_percentage, ctx.rank_score), ("情感倾向", ctx.sentiment_percentage, ctx.sentiment_score), ("引用质量", ctx.citation_percentage, ctx.citation_score), ("竞品对比", ctx.competitive_percentage, ctx.competitive_score), ] return sorted(dimensions, key=lambda x: x[1]) def _generate_content_optimization_suggestions( ctx: BrandAnalysisContext, ) -> list[SuggestionItem]: """生成内容优化建议""" suggestions: list[SuggestionItem] = [] # 提及率低 -> 内容优化 if ctx.mention_rate_percentage < 50: suggestions.append(SuggestionItem( type="content_optimization", priority="high", title="提升品牌内容在AI平台的可见性", description=( f"当前提及率仅{ctx.mention_rate_percentage:.0f}%，" f"品牌在AI回答中被提及的频率较低。" f"AI平台倾向于引用结构化、权威性强的内容。" ), action=( "1. 在官网和核心页面添加FAQ结构化数据(Schema.org FAQPage)\n" "2. 创建行业白皮书和深度分析文章，使用清晰的标题层级\n" "3. 确保品牌在维基百科、行业百科等权威来源有准确条目\n" "4. 优化内容中的品牌名称一致性，避免过多别名导致AI无法识别" ), expected_impact="预计可将提及率提升15-25个百分点", difficulty="medium", )) elif ctx.mention_rate_percentage < 75: suggestions.append(SuggestionItem( type="content_optimization", priority="medium", title="持续优化品牌内容结构", description=( f"当前提及率为{ctx.mention_rate_percentage:.0f}%，仍有提升空间。" f"建议进一步优化内容结构以提高AI引用概率。" ), action=( "1. 增加品牌相关的高质量长文内容（2000字以上）\n" "2. 在内容中增加数据支撑和案例引用\n" "3. 定期更新内容以保持时效性" ), expected_impact="预计可将提及率提升5-15个百分点", difficulty="easy", )) return suggestions def _generate_platform_targeting_suggestions( ctx: BrandAnalysisContext, ) -> list[SuggestionItem]: """生成平台定向建议""" suggestions: list[SuggestionItem] = [] # 找出评分最低的平台 if ctx.platform_scores: weak_platforms = sorted( ctx.platform_scores.items(), key=lambda x: x[1], )[:3] weak_platform_names = [p[0] for p in weak_platforms if p[1] < 40] weak_platform_str = "、".join(weak_platform_names) if weak_platform_names else "" if weak_platform_str: suggestions.append(SuggestionItem( type="platform_targeting", priority="high", title=f"重点优化{weak_platform_str}平台表现", description=( f"在这些平台上品牌评分低于40分，" f"AI引用率极低。不同AI平台有不同的内容偏好，" f"需要针对性优化。" ), action=( f"1. 分析{weak_platform_str}平台的内容偏好和引用模式\n" "2. 针对各平台优化内容格式和表达方式\n" "3. 增加在这些平台关联的内容源数量\n" "4. 关注平台算法更新，及时调整优化策略" ), expected_impact="预计可将弱平台评分提升20-30分", difficulty="hard", )) # 如果没有平台数据 if not ctx.platform_scores or all(v == 0 for v in ctx.platform_scores.values()): suggestions.append(SuggestionItem( type="platform_targeting", priority="high", title="启动全平台品牌监控", description=( "当前没有任何平台的引用数据。" "需要先在各AI平台建立品牌存在感。" ), action=( "1. 确保已添加足够的查询词覆盖核心业务关键词\n" "2. 等待系统完成首轮数据采集（通常需要1-2天）\n" "3. 采集完成后查看各平台评分，确定优先优化方向" ), expected_impact="获取全平台基准数据，为后续优化提供方向", difficulty="easy", )) return suggestions def _generate_competitor_gap_suggestions( ctx: BrandAnalysisContext, ) -> list[SuggestionItem]: """生成竞品差距建议""" suggestions: list[SuggestionItem] = [] # 竞品对比维度弱 if ctx.competitive_percentage < 40: # 找出领先的品牌 ahead_competitors = [] behind_competitors = [] if ctx.competitor_data: brand_mentions = ctx.competitor_data.get("brand_mentions", 0) for name, count in ctx.competitor_data.get("competitor_mentions", {}).items(): if count > brand_mentions: ahead_competitors.append((name, count)) else: behind_competitors.append((name, count)) ahead_str = "、".join([n for n, _ in ahead_competitors[:3]]) if ahead_competitors else "竞品" suggestions.append(SuggestionItem( type="competitor_gap", priority="high", title=f"缩小与{ahead_str}的差距", description=( f"当前竞品对比得分率仅{ctx.competitive_percentage:.0f}%，" f"品牌在AI引用中落后于主要竞品。" f"需要分析竞品的优势领域并制定追赶策略。" ), action=( "1. 分析竞品在AI平台被引用的内容类型和话题\n" "2. 找出竞品有而品牌缺失的内容领域\n" "3. 针对性创建竞品优势领域的优质内容\n" "4. 加强品牌差异化定位，突出独特价值主张" ), expected_impact="预计3-6个月内可将竞品对比得分率提升15-25个百分点", difficulty="hard", )) elif ctx.competitive_percentage < 70: suggestions.append(SuggestionItem( type="competitor_gap", priority="medium", title="持续巩固竞争优势", description=( f"当前竞品对比得分率为{ctx.competitive_percentage:.0f}%，" f"品牌处于中等水平，需要持续巩固并扩大优势。" ), action=( "1. 定期监控竞品动态和内容更新\n" "2. 在品牌优势领域持续输出高质量内容\n" "3. 关注新兴话题和趋势，抢占先机" ), expected_impact="预计可将竞品对比得分率提升10-15个百分点", difficulty="medium", )) return suggestions def _generate_query_expansion_suggestions( ctx: BrandAnalysisContext, ) -> list[SuggestionItem]: """生成查询词扩展建议""" suggestions: list[SuggestionItem] = [] # 查询词数量不足 if ctx.total_queries < 10: suggestions.append(SuggestionItem( type="query_expansion", priority="high" if ctx.total_queries < 3 else "medium", title="扩展查询词覆盖范围", description=( f"当前仅有{ctx.total_queries}个查询词，" f"覆盖范围不足，无法全面反映品牌在AI搜索中的表现。" f"更多查询词意味着更全面的品牌认知度画像。" ), action=( "1. 添加行业核心关键词（如：'XX行业推荐'、'XX解决方案'）\n" "2. 添加品牌相关长尾词（如：'XX品牌怎么样'、'XX vs 竞品'）\n" "3. 添加场景化查询词（如：'XX场景下选什么产品'）\n" "4. 建议至少添加10-20个查询词以获得可靠的分析结果" ), expected_impact="更多查询词可提升评分准确度，发现更多优化机会", difficulty="easy", )) # 提及率低且查询词不少 -> 可能需要优化查询词质量 elif ctx.total_queries >= 10 and ctx.mention_rate_percentage < 50: suggestions.append(SuggestionItem( type="query_expansion", priority="medium", title="优化查询词质量和相关性", description=( f"已有{ctx.total_queries}个查询词，但提及率仅{ctx.mention_rate_percentage:.0f}%。" f"可能是查询词与品牌核心业务关联度不够。" ), action=( "1. 检查现有查询词是否覆盖品牌核心业务场景\n" "2. 添加品牌擅长的专业领域相关查询词\n" "3. 删除与品牌无关的宽泛查询词\n" "4. 增加品牌独特卖点和差异化优势相关的查询词" ), expected_impact="优化查询词后可提升提及率10-20个百分点", difficulty="easy", )) return suggestions def _generate_citation_improvement_suggestions( ctx: BrandAnalysisContext, ) -> list[SuggestionItem]: """生成引用改善建议""" suggestions: list[SuggestionItem] = [] # 引用质量低 if ctx.citation_percentage < 40: suggestions.append(SuggestionItem( type="citation_improvement", priority="high", title="提升AI引用内容的质量和深度", description=( f"当前引用质量得分率仅{ctx.citation_percentage:.0f}%，" f"AI对品牌的引用多为浅层提及，缺乏深度正面描述。" f"高质量引用能显著影响用户决策。" ), action=( "1. 创建详细的产品对比页面，包含数据表格和评测结论\n" "2. 发布客户案例和成功故事，提供具体数据支撑\n" "3. 在内容中增加可引用的数据点和统计信息\n" "4. 优化内容结构，使用AI易于提取的格式（列表、表格、要点）" ), expected_impact="预计可将引用质量得分率提升15-25个百分点", difficulty="medium", )) # 推荐排名低 if ctx.rank_percentage < 40: suggestions.append(SuggestionItem( type="citation_improvement", priority="high", title="提升品牌在AI推荐中的排名位置", description=( f"当前推荐排名得分率仅{ctx.rank_percentage:.0f}%，" f"品牌在AI推荐列表中排名靠后，用户看到概率低。" f"排名越靠前，被用户选择的可能性越大。" ), action=( "1. 增加品牌在权威第三方平台的正面评价和推荐\n" "2. 优化品牌在行业榜单和评测中的排名\n" "3. 创建'最佳XX'、'XX推荐'类内容，增加被推荐概率\n" "4. 提升品牌官网的SEO表现，增加AI爬取到的概率" ), expected_impact="预计可将推荐排名提升2-3位", difficulty="medium", )) # 情感倾向差 negative_rate = 0.0 total_sentiment = sum(ctx.sentiment_data.values()) if total_sentiment > 0: negative_rate = ctx.sentiment_data.get("negative", 0) / total_sentiment if negative_rate > 0.3: suggestions.append(SuggestionItem( type="citation_improvement", priority="high", title="改善AI平台对品牌的负面评价", description=( f"当前负面评价占比{negative_rate:.0%}，" f"AI在引用品牌时倾向使用负面表述。" f"负面引用会严重影响用户对品牌的印象。" ), action=( "1. 分析负面引用的具体内容，找出主要批评点\n" "2. 针对性改进产品或服务，解决用户痛点\n" "3. 主动发布正面内容，稀释负面信息的影响\n" "4. 在官方渠道积极回应用户反馈和投诉" ), expected_impact="减少负面引用比例10-20个百分点", difficulty="hard", )) return suggestions def generate_rule_based_suggestions( ctx: BrandAnalysisContext, ) -> list[SuggestionItem]: """ 基于规则生成优化建议（不依赖LLM） 分析品牌数据差距，按照优先级生成可操作的建议。 """ all_suggestions: list[SuggestionItem] = [] # 收集各类建议 all_suggestions.extend(_generate_content_optimization_suggestions(ctx)) all_suggestions.extend(_generate_platform_targeting_suggestions(ctx)) all_suggestions.extend(_generate_competitor_gap_suggestions(ctx)) all_suggestions.extend(_generate_query_expansion_suggestions(ctx)) all_suggestions.extend(_generate_citation_improvement_suggestions(ctx)) # 按优先级排序: high > medium > low priority_order = {"high": 0, "medium": 1, "low": 2} all_suggestions.sort(key=lambda s: priority_order.get(s.priority, 1)) # 限制最多5条建议 return all_suggestions[:5] # ============================================================ # LLM 建议生成器 # ============================================================ OPTIMIZATION_PROMPT = """你是一个GEO（生成式引擎优化）专家。基于以下品牌数据，提供3-5条可操作的优化建议。 品牌: {brand_name} 当前评分: {overall_score}/100 评分维度: - 提及率: {mention_rate_score}/{mention_rate_max} ({mention_rate_percentage}%) - 推荐排名: {rank_score}/{rank_max} ({rank_percentage}%) - 情感倾向: {sentiment_score}/{sentiment_max} ({sentiment_percentage}%) - 引用质量: {citation_score}/{citation_max} ({citation_percentage}%) - 竞品对比: {competitive_score}/{competitive_max} ({competitive_percentage}%) 竞品对比数据: {competitor_data_str} 情感分析: {sentiment_data_str} 平台评分: {platform_data_str} 请返回JSON格式: {{ "suggestions": [ {{ "type": "content_optimization" | "platform_targeting" | "competitor_gap" | "query_expansion" | "citation_improvement", "priority": "high" | "medium" | "low", "title": "建议标题", "description": "详细描述", "action": "具体操作步骤", "expected_impact": "预期效果", "difficulty": "easy" | "medium" | "hard" }} ] }} 要求: 1. 每条建议必须基于数据，指出具体的差距和改进方向 2. 优先关注最弱的维度 3. 建议必须可操作，包含具体步骤 4. 预期效果要量化 5. 返回3-5条建议，按优先级从高到低排列 """ async def generate_llm_suggestions( ctx: BrandAnalysisContext, ) -> list[SuggestionItem]: """ 使用DeepSeek LLM生成个性化优化建议 如果LLM不可用或调用失败，回退到规则生成。 """ if not settings.ENABLE_LLM or not settings.DEEPSEEK_API_KEY: logger.info("LLM未启用或API Key未配置，使用规则生成建议") return generate_rule_based_suggestions(ctx) try: # 构建prompt competitor_data_str = json.dumps(ctx.competitor_data, ensure_ascii=False, indent=2) sentiment_data_str = json.dumps(ctx.sentiment_data, ensure_ascii=False, indent=2) platform_data_str = json.dumps(ctx.platform_scores, ensure_ascii=False, indent=2) prompt = OPTIMIZATION_PROMPT.format( brand_name=ctx.brand_name, overall_score=ctx.overall_score, mention_rate_score=round(ctx.mention_rate_score, 2), mention_rate_max=ctx.mention_rate_max, mention_rate_percentage=round(ctx.mention_rate_percentage, 1), rank_score=round(ctx.rank_score, 2), rank_max=ctx.rank_max, rank_percentage=round(ctx.rank_percentage, 1), sentiment_score=round(ctx.sentiment_score, 2), sentiment_max=ctx.sentiment_max, sentiment_percentage=round(ctx.sentiment_percentage, 1), citation_score=round(ctx.citation_score, 2), citation_max=ctx.citation_max, citation_percentage=round(ctx.citation_percentage, 1), competitive_score=round(ctx.competitive_score, 2), competitive_max=ctx.competitive_max, competitive_percentage=round(ctx.competitive_percentage, 1), competitor_data_str=competitor_data_str, sentiment_data_str=sentiment_data_str, platform_data_str=platform_data_str, ) # 调用DeepSeek API from openai import OpenAI client = OpenAI( api_key=settings.DEEPSEEK_API_KEY, base_url="https://api.deepseek.com", ) response = await asyncio.to_thread( client.chat.completions.create, model="deepseek-chat", messages=[{"role": "user", "content": prompt}], temperature=0.3, max_tokens=2000, ) content = response.choices[0].message.content if not content: raise ValueError("LLM返回空响应") # 提取JSON json_str = extract_json(content) result = json.loads(json_str) # 解析建议 suggestions: list[SuggestionItem] = [] for item in result.get("suggestions", []): # 验证type字段 valid_types = { "content_optimization", "platform_targeting", "competitor_gap", "query_expansion", "citation_improvement", } suggestion_type = item.get("type", "content_optimization") if suggestion_type not in valid_types: suggestion_type = "content_optimization" # 验证priority字段 valid_priorities = {"high", "medium", "low"} priority = item.get("priority", "medium") if priority not in valid_priorities: priority = "medium" # 验证difficulty字段 valid_difficulties = {"easy", "medium", "hard"} difficulty = item.get("difficulty", "medium") if difficulty not in valid_difficulties: difficulty = "medium" suggestions.append(SuggestionItem( type=suggestion_type, priority=priority, title=item.get("title", "优化建议"), description=item.get("description", ""), action=item.get("action", ""), expected_impact=item.get("expected_impact", ""), difficulty=difficulty, )) if not suggestions: logger.warning("LLM未返回有效建议，回退到规则生成") return generate_rule_based_suggestions(ctx) return suggestions[:5] except Exception as e: logger.error(f"LLM生成建议失败: {e}，回退到规则生成") return generate_rule_based_suggestions(ctx) # ============================================================ # 主入口：生成优化建议 # ============================================================ async def generate_suggestions( ctx: BrandAnalysisContext, ) -> list[SuggestionItem]: """ 生成优化建议 如果ENABLE_LLM=True且有DeepSeek API Key，使用LLM生成个性化建议。 否则使用基于规则的模板建议。 """ if settings.ENABLE_LLM and settings.DEEPSEEK_API_KEY: return await generate_llm_suggestions(ctx) return generate_rule_based_suggestions(ctx) def build_context_from_scoring_result( brand_name: str, scoring_result: ScoringResultV2, competitor_data: dict[str, Any] | None = None, sentiment_data: dict[str, int] | None = None, platform_scores: dict[str, float] | None = None, total_queries: int = 0, mentioned_count: int = 0, ) -> BrandAnalysisContext: """ 从ScoringResultV2构建BrandAnalysisContext Args: brand_name: 品牌名称 scoring_result: V2评分结果 competitor_data: 竞品对比数据 sentiment_data: 情感分析数据 platform_scores: 平台评分数据 total_queries: 总查询次数 mentioned_count: 被提及次数 Returns: BrandAnalysisContext: 品牌分析上下文 """ return BrandAnalysisContext( brand_name=brand_name, overall_score=scoring_result.overall_score, mention_rate_score=scoring_result.mention_rate.score, mention_rate_max=scoring_result.mention_rate.max_score, mention_rate_percentage=scoring_result.mention_rate.percentage, rank_score=scoring_result.recommendation_rank.score, rank_max=scoring_result.recommendation_rank.max_score, rank_percentage=scoring_result.recommendation_rank.percentage, sentiment_score=scoring_result.sentiment_score.score, sentiment_max=scoring_result.sentiment_score.max_score, sentiment_percentage=scoring_result.sentiment_score.percentage, citation_score=scoring_result.citation_quality.score, citation_max=scoring_result.citation_quality.max_score, citation_percentage=scoring_result.citation_quality.percentage, competitive_score=scoring_result.competitive_position.score, competitive_max=scoring_result.competitive_position.max_score, competitive_percentage=scoring_result.competitive_position.percentage, competitor_data=competitor_data or {}, sentiment_data=sentiment_data or {}, platform_scores=platform_scores or {}, total_queries=total_queries, mentioned_count=mentioned_count, )