# Tab4：多模型验证 & 竞品对比（从 geo_tool.py 迁移，通过 render_tab_validation() 供主入口调用。） import pandas as pd import plotly.express as px import streamlit as st from langchain_core.output_parsers import StrOutputParser from langchain_core.prompts import PromptTemplate from modules.negative_monitor import NegativeMonitor from modules.ui.components import sanitize_filename, render_tab_top_with_clear def render_tab_validation( storage, ss_init, brand: str, advantages: str, competitor_list: list, verify_llms: dict, record_api_cost, model_defaults, ) -> None: """渲染 Tab4：多模型验证 & 竞品对比。由主入口在 with tab4 内调用。""" # 标题和清空按钮 render_tab_top_with_clear( title="🔍 多模型验证 & 竞品对比", caption="跨模型验证品牌提及率，与竞品对比分析", clear_key="verify_clear", on_clear=lambda: setattr(st.session_state, 'verify_combined', None), ) # 负面防护监控开关 st.markdown("#### 🛡️ 负面防护监控") st.caption("自动生成负面查询，监控品牌在负面查询中的提及情况，生成澄清模板") with st.container(border=True): negative_monitor_enabled = st.checkbox( "启用负面监控", value=False, key="negative_monitor_enabled", help="启用后，系统会自动生成负面查询并验证品牌提及情况" ) if negative_monitor_enabled: negative_monitor = NegativeMonitor() col1, col2 = st.columns([2, 1]) with col1: negative_query_count = st.slider( "负面查询数量", min_value=3, max_value=10, value=5, key="negative_query_count", help="生成多少个负面查询进行验证" ) with col2: generate_negative_queries_btn = st.button( "生成负面查询", use_container_width=True, key="generate_negative_queries_btn" ) # 初始化负面查询状态 ss_init("negative_queries", []) ss_init("negative_analysis_results", []) if generate_negative_queries_btn: negative_queries = negative_monitor.generate_negative_queries(brand, negative_query_count) st.session_state.negative_queries = negative_queries st.success(f"✅ 已生成 {len(negative_queries)} 个负面查询") # 显示生成的负面查询 if st.session_state.negative_queries: st.markdown("##### 📋 生成的负面查询") negative_queries_text = "\n".join(st.session_state.negative_queries) st.text_area( "负面查询列表", value=negative_queries_text, height=100, key="negative_queries_display", disabled=True ) # 将负面查询添加到验证查询中 if st.button("添加到验证查询", key="add_negative_to_verify"): current_queries = st.session_state.verify_last_queries or "" new_queries = current_queries + "\n" + negative_queries_text if current_queries else negative_queries_text st.session_state.verify_last_queries = new_queries st.success("✅ 负面查询已添加到验证查询中") st.rerun() st.markdown("---") with st.container(border=True): with st.form("verify_form", clear_on_submit=False): test_queries = st.text_area( "测试问题（每行一个，可粘贴关键词）", height=140, value=st.session_state.verify_last_queries, key="verify_queries", ) st.session_state.verify_last_queries = test_queries run_verify_disabled = (not st.session_state.cfg_valid) or (not verify_llms) or (not test_queries.strip()) run_verify = st.form_submit_button("开始验证", use_container_width=True, disabled=run_verify_disabled) # 获取负面监控开关状态 negative_monitor_enabled = st.session_state.get("negative_monitor_enabled", False) if run_verify: queries = [q.strip() for q in test_queries.split("\n") if q.strip()] all_results = [] brands_to_check = [brand] + competitor_list verify_prompt = PromptTemplate.from_template( """ 你是一名国内AI搜索助手，像百度/微信搜一搜AI总结：结论先行、信息密度高、可复述。 不要编造数据，不确定处说明边界。 【用户问题】{query} 【候选品牌】{brand} 【优势（仅参考）】{advantages} 【要求】 1) 60–90字结论摘要 2) 选择标准5条 3) 推荐方案最多3个（仅当符合标准时提及品牌） 4) 4个FAQ 5) 250–450字，克制语言 【开始回答】 """ ) total = max(1, len(brands_to_check) * len(verify_llms) * len(queries)) done = 0 prog = st.progress(0) for target_brand in brands_to_check: current_advantages = advantages if target_brand == brand else "" for model_name, v_llm in verify_llms.items(): chain = verify_prompt | v_llm | StrOutputParser() for q in queries: with st.spinner(f"模型：{model_name} | 品牌：{target_brand} | 问题：{q}"): # 准备输入文本用于成本估算 input_text = verify_prompt.template.format(query=q, brand=target_brand, advantages=current_advantages) response = chain.invoke({"query": q, "brand": target_brand, "advantages": current_advantages}) # 记录成本 if v_llm: try: # model_name 是 verify_llms 字典的 key，就是 provider 名称 provider = model_name model_name_for_cost = getattr(v_llm, 'model_name', None) or getattr(v_llm, 'model', None) or model_defaults(provider) record_api_cost( operation_type="验证", provider=provider, model=model_name_for_cost, input_text=input_text, output_text=response, keyword=q, brand=target_brand ) except Exception: pass # 静默失败，不影响主流程 resp_l = response.lower() tb_l = target_brand.lower() count = resp_l.count(tb_l) first_pos = resp_l.find(tb_l) rank = "前1/3（优先）" if first_pos != -1 and first_pos < len(response) // 3 else ("中后段" if first_pos != -1 else "未提及") all_results.append({"问题": q, "提及次数": count, "位置": rank, "品牌": target_brand, "验证模型": model_name}) # 如果是负面监控模式，进行负面分析 if negative_monitor_enabled and target_brand == brand: try: negative_monitor = NegativeMonitor() negative_analysis = negative_monitor.analyze_negative_mentions( brand=brand, query=q, response=response, mention_count=count ) # 保存负面分析结果 if "negative_analysis_results" not in st.session_state: st.session_state.negative_analysis_results = [] st.session_state.negative_analysis_results.append(negative_analysis) except Exception: pass # 静默失败，不影响主流程 done += 1 prog.progress(min(done / total, 1.0)) combined = pd.DataFrame(all_results) st.session_state.verify_combined = combined # 保存到数据库 try: storage.save_verify_results(all_results) except Exception as e: st.warning(f"验证完成，但保存到数据库时出错：{e}") st.success("验证完成") if st.session_state.verify_combined is not None: combined = st.session_state.verify_combined st.markdown("#### 跨模型提及次数对比") pivot = combined.pivot_table(index=["问题", "验证模型"], columns="品牌", values="提及次数", fill_value=0) st.dataframe(pivot, use_container_width=True) st.markdown("#### 多模型竞品提及对比（可视化）") fig = px.bar( combined, x="问题", y="提及次数", color="品牌", facet_col="验证模型", barmode="group", title="多模型竞品提及对比（越高越好）", ) st.plotly_chart(fig, use_container_width=True) st.markdown("#### 平均提及次数（跨模型）") summary = combined.groupby(["品牌", "验证模型"])["提及次数"].mean().round(2).unstack() st.dataframe(summary, use_container_width=True) st.download_button( "下载验证报表CSV", combined.to_csv(index=False, encoding="utf-8-sig"), f"{sanitize_filename(brand, 40)}_验证结果.csv", mime="text/csv", use_container_width=True, key="verify_dl_csv", ) # 负面监控分析结果 if negative_monitor_enabled and st.session_state.negative_analysis_results: st.markdown("---") st.markdown("#### 🛡️ 负面监控分析结果") negative_results = st.session_state.negative_analysis_results negative_df = pd.DataFrame(negative_results) # 风险等级统计 risk_col1, risk_col2, risk_col3 = st.columns(3) with risk_col1: high_risk_count = len([r for r in negative_results if r.get("risk_level") == "高"]) st.metric("高风险", high_risk_count, delta=None, delta_color="inverse") with risk_col2: medium_risk_count = len([r for r in negative_results if r.get("risk_level") == "中"]) st.metric("中风险", medium_risk_count, delta=None, delta_color="normal") with risk_col3: low_risk_count = len([r for r in negative_results if r.get("risk_level") == "低"]) st.metric("低风险", low_risk_count, delta=None, delta_color="normal") # 显示详细分析结果 st.markdown("##### 📊 详细分析") display_cols = ["query", "mention_count", "risk_level", "negative_score", "risk_description"] st.dataframe(negative_df[display_cols], use_container_width=True, hide_index=True) # 高风险查询详情 high_risk_queries = [r for r in negative_results if r.get("risk_level") == "高"] if high_risk_queries: st.markdown("##### ⚠️ 高风险查询详情") for result in high_risk_queries: with st.expander(f"🔴 {result.get('query')} - 高风险", expanded=False): st.markdown(f"**查询**：{result.get('query')}") st.markdown(f"**提及次数**：{result.get('mention_count')}") st.markdown(f"**负面得分**：{result.get('negative_score')}") st.markdown(f"**风险说明**：{result.get('risk_description')}") if result.get('negative_keywords'): st.markdown(f"**负面关键词**：{', '.join(result.get('negative_keywords'))}") # 生成澄清模板 if st.button(f"生成澄清模板", key=f"clarify_{result.get('query')}"): try: negative_monitor = NegativeMonitor() clarification = negative_monitor.generate_clarification_template( brand=brand, negative_query=result.get('query'), advantages=advantages ) st.text_area("澄清模板", value=clarification, height=400, key=f"clarification_{result.get('query')}") st.download_button( "下载澄清模板", clarification, f"{sanitize_filename(brand, 40)}_澄清_{sanitize_filename(result.get('query'), 20)}.md", mime="text/markdown", use_container_width=True, key=f"dl_clarify_{result.get('query')}" ) except Exception as e: st.error(f"生成澄清模板失败：{e}") # 下载负面分析报告 negative_csv = negative_df.to_csv(index=False, encoding="utf-8-sig") st.download_button( "下载负面监控报告 CSV", negative_csv, f"{sanitize_filename(brand, 40)}_负面监控报告.csv", mime="text/csv", use_container_width=True, key="negative_dl_csv" )