|
|
@@ -0,0 +1,669 @@
|
|
|
+{
|
|
|
+ "cells": [
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "38b007c8",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "\n",
|
|
|
+ "## GiftGeniusAgent——你的送礼智能Agent\n",
|
|
|
+ "\n",
|
|
|
+ "### 项目简介\n",
|
|
|
+ "本项目演示一个基于HelloAgents框架的智能送礼Agent\n",
|
|
|
+ "\n",
|
|
|
+ "### 作者信息\n",
|
|
|
+ "- 姓名：张善祺\n",
|
|
|
+ "- GitHub:@jack6249\n",
|
|
|
+ "- 日期：2025-11-21\n"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "62a3b6a3",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "### 第1部分：环境配置"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "code",
|
|
|
+ "execution_count": null,
|
|
|
+ "id": "dd9b8a20",
|
|
|
+ "metadata": {},
|
|
|
+ "outputs": [],
|
|
|
+ "source": [
|
|
|
+ "#导入库和参数配置\n",
|
|
|
+ "from hello_agents import SimpleAgent, HelloAgentsLLM, ReflectionAgent\n",
|
|
|
+ "from hello_agents.tools import Tool, ToolParameter\n",
|
|
|
+ "from typing import Dict, Any, List\n",
|
|
|
+ "from tavily import TavilyClient\n",
|
|
|
+ "import os\n",
|
|
|
+ "import json\n",
|
|
|
+ "import re\n",
|
|
|
+ "import numpy as np \n",
|
|
|
+ "from dotenv import load_dotenv\n",
|
|
|
+ "\n",
|
|
|
+ "load_dotenv()\n",
|
|
|
+ "\n",
|
|
|
+ "#LLM参数\n",
|
|
|
+ "LLM_MODEL_ID = os.getenv(\"LLM_MODEL_ID\")\n",
|
|
|
+ "#\"deepseek-chat\"\n",
|
|
|
+ "LLM_API_KEY = os.getenv(\"LLM_API_KEY\")\n",
|
|
|
+ "LLM_BASE_URL = os.getenv(\"LLM_BASE_URL\")\n",
|
|
|
+ "#Tavily参数\n",
|
|
|
+ "TAVILY_API_KEY = os.getenv(\"TAVILY_API_KEY\")\n",
|
|
|
+ "\n",
|
|
|
+ "print(\"✅ 环境配置完成\")\n"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "27482621",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "### 第2部分：定义工具"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "code",
|
|
|
+ "execution_count": null,
|
|
|
+ "id": "2087de45",
|
|
|
+ "metadata": {},
|
|
|
+ "outputs": [],
|
|
|
+ "source": [
|
|
|
+ "class BatchSearchTool(Tool):\n",
|
|
|
+ " def __init__(self):\n",
|
|
|
+ " super().__init__(\n",
|
|
|
+ " name=\"batch_search\",\n",
|
|
|
+ " description=\"高级批量搜索工具。\"\n",
|
|
|
+ " )\n",
|
|
|
+ "\n",
|
|
|
+ " def search_raw(self, query: str) -> List[Dict]:\n",
|
|
|
+ " if \"TAVILY_API_KEY\" not in os.environ: \n",
|
|
|
+ " print(\"❌ 错误：缺少 API Key\")\n",
|
|
|
+ " return []\n",
|
|
|
+ "\n",
|
|
|
+ " print(f\" 🚀 [直连搜索] 正在抓取: {query} ...\")\n",
|
|
|
+ " try:\n",
|
|
|
+ " tavily = TavilyClient(api_key=os.environ[\"TAVILY_API_KEY\"])\n",
|
|
|
+ " # 搜索包含图片\n",
|
|
|
+ " response = tavily.search(query, max_results=5, include_images=True) # 增加到5条，提高命中率\n",
|
|
|
+ " \n",
|
|
|
+ " results = []\n",
|
|
|
+ " # 1. 提取文本结果\n",
|
|
|
+ " if 'results' in response:\n",
|
|
|
+ " for r in response['results']:\n",
|
|
|
+ " results.append({\n",
|
|
|
+ " \"title\": r['title'],\n",
|
|
|
+ " \"url\": r['url'],\n",
|
|
|
+ " \"content\": r['content'], \n",
|
|
|
+ " \"type\": \"text\"\n",
|
|
|
+ " })\n",
|
|
|
+ " \n",
|
|
|
+ " # 2. 提取图片结果\n",
|
|
|
+ " if 'images' in response and response['images']:\n",
|
|
|
+ " results.append({\n",
|
|
|
+ " \"images\": response['images'][:3], # 取前3张\n",
|
|
|
+ " \"type\": \"image\"\n",
|
|
|
+ " })\n",
|
|
|
+ " \n",
|
|
|
+ " return results\n",
|
|
|
+ " \n",
|
|
|
+ " except Exception as e:\n",
|
|
|
+ " print(f\" ⚠️ 搜索异常: {e}\")\n",
|
|
|
+ " return []\n",
|
|
|
+ " \n",
|
|
|
+ " # 兼容 Agent 调用接口\n",
|
|
|
+ " def run(self, parameters: Any) -> str:\n",
|
|
|
+ " return \"请使用 Python 代码直接调用 search_raw 方法获取数据。\"\n",
|
|
|
+ " \n",
|
|
|
+ " def get_parameters(self) -> List[ToolParameter]:\n",
|
|
|
+ " return [ToolParameter(name=\"query\", type=\"string\", description=\"关键词\", required=True)]\n",
|
|
|
+ "\n",
|
|
|
+ "print(\"✅ BatchSearchTool定义完成\")"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "e8ffecb0",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "### 第3部分：创建智能体\n",
|
|
|
+ "\n"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "code",
|
|
|
+ "execution_count": null,
|
|
|
+ "id": "6c5a2e82",
|
|
|
+ "metadata": {},
|
|
|
+ "outputs": [],
|
|
|
+ "source": [
|
|
|
+ "from hello_agents import ToolRegistry\n",
|
|
|
+ "\n",
|
|
|
+ "#创建工具\n",
|
|
|
+ "tool_registry = ToolRegistry()\n",
|
|
|
+ "tool_registry.register_tool(BatchSearchTool())\n",
|
|
|
+ "\n",
|
|
|
+ "print(tool_registry.list_tools())\n",
|
|
|
+ "\n",
|
|
|
+ "\n",
|
|
|
+ "# 初始化大模型\n",
|
|
|
+ "llm = HelloAgentsLLM()\n",
|
|
|
+ "\n",
|
|
|
+ "# --- 1. 军师 (Profiler) - 已升级支持多维度画像 ---\n",
|
|
|
+ "PROFILER_PROMPT = \"\"\"\n",
|
|
|
+ "你是一个精通 MBTI 人格分析与消费市场趋势的 \"送礼军师\"。\n",
|
|
|
+ "你的任务是根据用户提供的多维度画像，制定 3 个**极度精准**的搜索关键词。\n",
|
|
|
+ "\n",
|
|
|
+ "【⚠️ 时效性死命令 (CRITICAL)】\n",
|
|
|
+ "当前时间视作 **2025年11月**。\n",
|
|
|
+ "1. **严禁过时**：绝对不要推荐 2024 年或更早的旧款（除非是经典恒久款如黑胶唱片）。\n",
|
|
|
+ "2. **锁定新品**：对于**化妆品、数码、盲盒**，必须搜索 **\"2025圣诞限定\"**、**\"2025秋冬新品\"** 或 **\"2026春季预告\"**。\n",
|
|
|
+ "3. **价格尺度**：给出的价格单位是人民币元。请严格遵照范围进行联想，禁止超出预算范围。\n",
|
|
|
+ "\n",
|
|
|
+ "为了确保推荐质量，请参考以下的【优秀思考范例】：\n",
|
|
|
+ "\n",
|
|
|
+ "### 范例 1\n",
|
|
|
+ "**用户画像**: \n",
|
|
|
+ "- 女, 26岁, ISFP (探险家), 金牛座\n",
|
|
|
+ "- 预算: 500-1000元\n",
|
|
|
+ "- 场景: 情人节\n",
|
|
|
+ "- 自定义: 喜欢有质感的生活小物\n",
|
|
|
+ "**军师分析**: \n",
|
|
|
+ "ISFP 重视审美和感官体验，金牛座喜欢实实在在的质感。情人节需要浪漫。\n",
|
|
|
+ "**生成策略**:\n",
|
|
|
+ "1. 观夏 (To Summer) 昆仑煮雪 晶石香薰 (符合质感与审美)\n",
|
|
|
+ "2. 野兽派 2025 情人节限定 睡衣礼盒 (金牛座喜欢的舒适)\n",
|
|
|
+ "3. 富士 Instax mini Evo 拍立得 (记录生活瞬间)\n",
|
|
|
+ "\n",
|
|
|
+ "### 范例 2\n",
|
|
|
+ "**用户画像**: \n",
|
|
|
+ "- 男, 30岁, INTJ (建筑师), 处女座\n",
|
|
|
+ "- 预算: 1000元以上\n",
|
|
|
+ "- 场景: 生日\n",
|
|
|
+ "- 自定义: 程序员，喜欢整洁\n",
|
|
|
+ "**军师分析**: \n",
|
|
|
+ "INTJ 追求极致的逻辑和效率，处女座有洁癖，喜欢桌面整洁。\n",
|
|
|
+ "**生成策略**:\n",
|
|
|
+ "1. Keychron Q1 Pro 机械键盘 铝坨坨 (符合极客对工具的追求)\n",
|
|
|
+ "2. 明基 (BenQ) ScreenBar Halo 屏幕挂灯 (极致护眼与桌面美学)\n",
|
|
|
+ "3. 赫曼米勒 (Herman Miller) 显示器支架 (人体工学)\n",
|
|
|
+ "\n",
|
|
|
+ "### 范例 3\n",
|
|
|
+ "**用户画像**: \n",
|
|
|
+ "- 女, 20岁, ENFP (竞选者), 狮子座\n",
|
|
|
+ "- 预算: 300元以内\n",
|
|
|
+ "- 场景: 圣诞节\n",
|
|
|
+ "- 自定义: 喜欢二次元，痛包\n",
|
|
|
+ "**军师分析**: \n",
|
|
|
+ "ENFP 热情奔放，狮子座喜欢张扬、闪亮的东西。预算有限但要素多。\n",
|
|
|
+ "**生成策略**:\n",
|
|
|
+ "1. 泡泡玛特 圣诞系列 盲盒整端 (符合节日气氛和二次元)\n",
|
|
|
+ "2. WEGO 痛包 镭射款 (符合自定义需求，狮子座喜欢的亮眼)\n",
|
|
|
+ "3. Chiikawa 吉伊卡哇 圣诞公仔 (当下顶流二次元IP)\n",
|
|
|
+ "\n",
|
|
|
+ "---\n",
|
|
|
+ "\n",
|
|
|
+ "**现在的任务**：\n",
|
|
|
+ "请根据以下【当前用户画像】进行分析，模仿上述范例的深度，制定搜索策略。\n",
|
|
|
+ "\n",
|
|
|
+ "【当前用户画像】\n",
|
|
|
+ "{user_profile_text}\n",
|
|
|
+ "\n",
|
|
|
+ "【关键词生成要求】\n",
|
|
|
+ "1. **必须具体**：格式为 `[品牌] + [产品名/系列] + [限定/属性]`。\n",
|
|
|
+ "2. **拒绝大词**：严禁搜索 \"礼物\"、\"口红\"、\"玩具\" 这种泛词。\n",
|
|
|
+ "3. **必须包含品牌**：根据预算推断合适的品牌（如：预算低选名创优品/泡泡玛特，预算高选Dior/索尼）。\n",
|
|
|
+ "\n",
|
|
|
+ "【输出格式】\n",
|
|
|
+ "只输出 3 行搜索关键词，每行一个。不要输出分析过程，不要序号。\n",
|
|
|
+ "\"\"\"\n",
|
|
|
+ "\n",
|
|
|
+ "profiler_agent = SimpleAgent(\n",
|
|
|
+ " llm=llm,\n",
|
|
|
+ " name=\"Agent_Profiler\",\n",
|
|
|
+ " system_prompt=PROFILER_PROMPT\n",
|
|
|
+ ")\n",
|
|
|
+ "\n",
|
|
|
+ "# ==============================================================================\n",
|
|
|
+ "# 2. 提取专家 (Extractor) - 数据清洗 (加入反思机制)\n",
|
|
|
+ "# ==============================================================================\n",
|
|
|
+ "EXTRACTOR_PROMPTS = {\n",
|
|
|
+ " \"initial\": \"\"\"\n",
|
|
|
+ " 你是一个 **纯粹的数据提取器**。\n",
|
|
|
+ " 你的任务是从 `<<<原始数据>>>` 中提取商品参数。\n",
|
|
|
+ "\n",
|
|
|
+ " ### 🧠 CoT 提取逻辑\n",
|
|
|
+ " 1. **扫描**: 快速浏览文本，寻找 \"¥\", \"$\", \"http\" 等关键符号。\n",
|
|
|
+ " 2. **验证**: 确认找到的信息是否属于\"核心商品\"（而非广告推荐）。\n",
|
|
|
+ " 3. **格式化**: 将提取到的信息填入 JSON。\n",
|
|
|
+ "\n",
|
|
|
+ " ### 🚫 绝对禁令\n",
|
|
|
+ " 1. **严禁伪造**: 没找到价格就填 \"暂无报价\"，绝对不要编数字！\n",
|
|
|
+ " 2. **严禁 API 格式**: 不要输出 `{{ \"status\": ... }}`。\n",
|
|
|
+ " 3. **严禁 Markdown**: 直接输出列表 `[...]`。\n",
|
|
|
+ "\n",
|
|
|
+ " ### ✅ 正确范例\n",
|
|
|
+ " 输入: ...Dior 口红...价格 ¥350...图片 http://img...\n",
|
|
|
+ " 输出:\n",
|
|
|
+ " [\n",
|
|
|
+ " {{\n",
|
|
|
+ " \"name\": \"Dior 口红\",\n",
|
|
|
+ " \"price\": \"¥350\",\n",
|
|
|
+ " \"img\": \"http://img...\"\n",
|
|
|
+ " }}\n",
|
|
|
+ " ]\n",
|
|
|
+ "\n",
|
|
|
+ " ---\n",
|
|
|
+ " <<<原始数据开始>>>\n",
|
|
|
+ " {task}\n",
|
|
|
+ " <<<原始数据结束>>>\n",
|
|
|
+ " \n",
|
|
|
+ " 请提取数据并输出 JSON 列表：\n",
|
|
|
+ " \"\"\",\n",
|
|
|
+ " \"reflect\": \"检查输出：是否以 `[` 开头？是否包含幻觉商品？\",\n",
|
|
|
+ " \"refine\": \"修正格式，只输出纯净的 JSON 列表 `[...]`。\"\n",
|
|
|
+ "}\n",
|
|
|
+ "\n",
|
|
|
+ "extractor_agent = ReflectionAgent(\n",
|
|
|
+ " name=\"Agent_Extractor\",\n",
|
|
|
+ " llm=llm,\n",
|
|
|
+ " max_iterations=1, \n",
|
|
|
+ " custom_prompts=EXTRACTOR_PROMPTS\n",
|
|
|
+ ")\n",
|
|
|
+ "\n",
|
|
|
+ "# ==============================================================================\n",
|
|
|
+ "# 3. 种草达人 (Pitcher) - 文案创作 (加入风格指导)\n",
|
|
|
+ "# ==============================================================================\n",
|
|
|
+ "PITCHER_PROMPT = \"\"\"\n",
|
|
|
+ "你是一个 **金牌种草文案**。\n",
|
|
|
+ "用户会给你一个 **【商品名称】**。\n",
|
|
|
+ "\n",
|
|
|
+ "### 🎯 关键要点 (Few Points)\n",
|
|
|
+ "1. **痛点直击**: 一句话说清楚为什么买它（限定？显白？绝美？）。\n",
|
|
|
+ "2. **情绪价值**: 使用 \"绝绝子\", \"氛围感\", \"心动\" 等高频热词。\n",
|
|
|
+ "3. **字数限制**: 严格控制在 **40字以内**，短小精悍。\n",
|
|
|
+ "4. **Emoji**: 必须包含 1-2 个 emoji。\n",
|
|
|
+ "\n",
|
|
|
+ "### 🌟 创作范例 (Few-Shot)\n",
|
|
|
+ "**输入**: Dior 999 烈艳蓝金\n",
|
|
|
+ "**输出**: 💄本宫不死终是妃！Dior 999 传奇正红，显白更有气场，送女友绝对没错！\n",
|
|
|
+ "\n",
|
|
|
+ "**输入**: 泡泡玛特 Labubu 坐坐派对\n",
|
|
|
+ "**输出**: ✨太可爱了吧！Labubu 坐坐派对系列，每一个都丑萌到心巴上，摆在桌上超治愈~\n",
|
|
|
+ "\n",
|
|
|
+ "**输入**: 罗技 MX Master 3S\n",
|
|
|
+ "**输出**: 🖱️打工人本命！罗技 Master 3S 静音又顺滑，人体工学设计，手腕再也不累了。\n",
|
|
|
+ "\n",
|
|
|
+ "---\n",
|
|
|
+ "\n",
|
|
|
+ "**当前任务**：\n",
|
|
|
+ "请为【{input}】写一句朋友圈风格种草语。\n",
|
|
|
+ "\"\"\"\n",
|
|
|
+ "pitcher_agent = SimpleAgent(\n",
|
|
|
+ " llm=llm,\n",
|
|
|
+ " name=\"Agent_Pitcher\",\n",
|
|
|
+ " system_prompt=PITCHER_PROMPT\n",
|
|
|
+ ")\n",
|
|
|
+ "\n",
|
|
|
+ "print(\"✅ 智能体初始化完成！\")"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "25907c14",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "### 第4部分:读取数据"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "code",
|
|
|
+ "execution_count": null,
|
|
|
+ "id": "58bd39dd",
|
|
|
+ "metadata": {},
|
|
|
+ "outputs": [],
|
|
|
+ "source": [
|
|
|
+ "\n",
|
|
|
+ "INPUT_FILENAME = \"data/test_cases.json\"\n",
|
|
|
+ "\n",
|
|
|
+ "def load_user_profile(filename):\n",
|
|
|
+ " # 1. 检查文件是否存在\n",
|
|
|
+ " if not os.path.exists(filename):\n",
|
|
|
+ " print(f\"⚠️ 未找到配置文件: {filename}\")\n",
|
|
|
+ " # 如果没有文件，将默认数据写入文件\n",
|
|
|
+ " default_data = {\n",
|
|
|
+ " \"性别\": \"女\",\n",
|
|
|
+ " \"年龄\": \"24岁\",\n",
|
|
|
+ " \"MBTI\": \"ENFP\",\n",
|
|
|
+ " \"星座\": \"天秤座\",\n",
|
|
|
+ " \"预算\": \"500元以内\",\n",
|
|
|
+ " \"节日\": \"恋爱一周年纪念日\",\n",
|
|
|
+ " \"自定义\": \"喜欢二次元，平时喜欢喝咖啡，不要送太实用的家电\"\n",
|
|
|
+ " }\n",
|
|
|
+ " with open(filename, \"w\", encoding=\"utf-8\") as f:\n",
|
|
|
+ " json.dump(default_data, f, ensure_ascii=False, indent=4)\n",
|
|
|
+ " print(f\"✅ 已自动生成默认配置文件，请修改 {filename} 后再次运行。\")\n",
|
|
|
+ " return default_data\n",
|
|
|
+ "\n",
|
|
|
+ " # 2. 读取文件内容\n",
|
|
|
+ " try:\n",
|
|
|
+ " with open(filename, \"r\", encoding=\"utf-8\") as f:\n",
|
|
|
+ " data = json.load(f)\n",
|
|
|
+ " print(f\"✅ 成功加载用户画像: {filename}\")\n",
|
|
|
+ " print(f\"📋 内容预览: {json.dumps(data, ensure_ascii=False)}\")\n",
|
|
|
+ " return data\n",
|
|
|
+ " except Exception as e:\n",
|
|
|
+ " print(f\"❌ 读取 JSON 失败: {e}\")\n",
|
|
|
+ " return {}\n",
|
|
|
+ "\n",
|
|
|
+ "# 加载数据\n",
|
|
|
+ "user_input_data = load_user_profile(INPUT_FILENAME)\n",
|
|
|
+ "\n",
|
|
|
+ "\n"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "0fdcab5a",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "### 第5部分：生成礼物计划"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "code",
|
|
|
+ "execution_count": null,
|
|
|
+ "id": "cdaef6b1",
|
|
|
+ "metadata": {},
|
|
|
+ "outputs": [],
|
|
|
+ "source": [
|
|
|
+ "\n",
|
|
|
+ "def parse_budget_range(budget_str):\n",
|
|
|
+ " \"\"\"解析用户预算字符串，返回 (min, max)\"\"\"\n",
|
|
|
+ " nums = [float(x) for x in re.findall(r'\\d+', str(budget_str).replace(',', ''))]\n",
|
|
|
+ " if not nums: return 0, 999999 \n",
|
|
|
+ " if \"以内\" in budget_str or \"以下\" in budget_str: return 0, nums[0]\n",
|
|
|
+ " if \"以上\" in budget_str: return nums[0], 999999\n",
|
|
|
+ " if len(nums) >= 2: return min(nums), max(nums)\n",
|
|
|
+ " return 0, nums[0]\n",
|
|
|
+ "\n",
|
|
|
+ "def extract_all_prices(raw_results):\n",
|
|
|
+ " \"\"\"从搜索结果列表中提取所有有效的价格\"\"\"\n",
|
|
|
+ " prices = []\n",
|
|
|
+ " for res in raw_results:\n",
|
|
|
+ " # 只处理文本类型的结果\n",
|
|
|
+ " if res.get('type') == 'text':\n",
|
|
|
+ " text = res.get('title', '') + \" \" + res.get('content', '')\n",
|
|
|
+ " # 匹配 ¥, $, 元 等格式\n",
|
|
|
+ " matches = re.findall(r'(?:¥|￥|\\$|HK\\$|NT\\$)\\s*(\\d+(?:,\\d{3})*(?:\\.\\d+)?)', text)\n",
|
|
|
+ " for m in matches:\n",
|
|
|
+ " val = float(m.replace(',', ''))\n",
|
|
|
+ " # 过滤掉像年份(2025)或过小/过大的异常值\n",
|
|
|
+ " if 10 < val < 100000 and val not in [2024, 2025, 2026]:\n",
|
|
|
+ " prices.append(val)\n",
|
|
|
+ " # 备用正则：匹配 \"xxx元\"\n",
|
|
|
+ " matches_yuan = re.findall(r'(\\d+(?:,\\d{3})*(?:\\.\\d+)?)\\s*元', text)\n",
|
|
|
+ " for m in matches_yuan:\n",
|
|
|
+ " val = float(m.replace(',', ''))\n",
|
|
|
+ " if 10 < val < 100000 and val not in [2024, 2025, 2026]:\n",
|
|
|
+ " prices.append(val)\n",
|
|
|
+ " return prices\n",
|
|
|
+ "\n",
|
|
|
+ "def find_best_product(hunter, profiler_agent, keyword, budget_limit):\n",
|
|
|
+ " \"\"\"\n",
|
|
|
+ " 智能搜索核心函数：包含 初次搜索 -> 价格校验 -> 策略修正 -> 自动降级 -> 智能兜底\n",
|
|
|
+ " \"\"\"\n",
|
|
|
+ " # 容器：用于收集所有搜索到的潜在结果（用于兜底）\n",
|
|
|
+ " all_candidates = []\n",
|
|
|
+ " current_kw = keyword\n",
|
|
|
+ " \n",
|
|
|
+ " # --- Round 1: 首次搜索 ---\n",
|
|
|
+ " print(f\" 🕵️ 第1次搜索: {current_kw} 价格\")\n",
|
|
|
+ " results_1 = hunter.search_raw(f\"{current_kw} 价格 RMB\")\n",
|
|
|
+ " \n",
|
|
|
+ " # 提取第一轮搜索到的通用图片（以防具体条目没图）\n",
|
|
|
+ " fallback_img = \"\"\n",
|
|
|
+ " for r in results_1:\n",
|
|
|
+ " if r.get('images'): \n",
|
|
|
+ " fallback_img = r['images'][0]\n",
|
|
|
+ " break\n",
|
|
|
+ "\n",
|
|
|
+ " # 检查 Round 1 结果\n",
|
|
|
+ " for res in results_1:\n",
|
|
|
+ " if res.get('type') == 'text':\n",
|
|
|
+ " p_vals = extract_all_prices([res])\n",
|
|
|
+ " if p_vals:\n",
|
|
|
+ " res['price_val'] = p_vals[0]\n",
|
|
|
+ " all_candidates.append(res) # 存入候选池\n",
|
|
|
+ " if p_vals[0] <= budget_limit:\n",
|
|
|
+ " # 完美命中！\n",
|
|
|
+ " if not res.get('img') and fallback_img: res['img'] = fallback_img\n",
|
|
|
+ " return res, f\"约 {p_vals[0]}元\", current_kw\n",
|
|
|
+ "\n",
|
|
|
+ " # --- Round 2: 策略修正 (Feedback Loop) ---\n",
|
|
|
+ " # 只有当找到了价格但都超标时，才触发修正\n",
|
|
|
+ " avg_price = np.mean([c['price_val'] for c in all_candidates]) if all_candidates else 0\n",
|
|
|
+ " \n",
|
|
|
+ " if avg_price > budget_limit:\n",
|
|
|
+ " print(f\" 💸 价格超标 (均价 {int(avg_price)} > 预算 {int(budget_limit/1.2)})，触发修正...\")\n",
|
|
|
+ " \n",
|
|
|
+ " correction_prompt = f\"\"\"\n",
|
|
|
+ " 原策略 \"{current_kw}\" 市场均价约 {int(avg_price)}元，超预算。\n",
|
|
|
+ " 请推荐一个 **同品类但更便宜** 的具体型号（平替）。\n",
|
|
|
+ " 只输出关键词，不要解释。\n",
|
|
|
+ " \"\"\"\n",
|
|
|
+ " new_kw = profiler_agent.run(correction_prompt).strip()\n",
|
|
|
+ " print(f\" 🔄 军师修正策略为: {new_kw}\")\n",
|
|
|
+ " current_kw = new_kw # 更新关键词\n",
|
|
|
+ " \n",
|
|
|
+ " # 执行第二次搜索\n",
|
|
|
+ " results_2 = hunter.search_raw(f\"{new_kw} 价格 RMB\")\n",
|
|
|
+ " \n",
|
|
|
+ " # 提取第二轮的图片\n",
|
|
|
+ " for r in results_2:\n",
|
|
|
+ " if r.get('images'): \n",
|
|
|
+ " fallback_img = r['images'][0]\n",
|
|
|
+ " break\n",
|
|
|
+ " \n",
|
|
|
+ " # 检查 Round 2 结果\n",
|
|
|
+ " for res in results_2:\n",
|
|
|
+ " if res.get('type') == 'text':\n",
|
|
|
+ " p_vals = extract_all_prices([res])\n",
|
|
|
+ " if p_vals:\n",
|
|
|
+ " res['price_val'] = p_vals[0]\n",
|
|
|
+ " all_candidates.append(res) \n",
|
|
|
+ " if p_vals[0] <= budget_limit:\n",
|
|
|
+ " if not res.get('img') and fallback_img: res['img'] = fallback_img\n",
|
|
|
+ " return res, f\"约 {p_vals[0]}元 (平替)\", current_kw\n",
|
|
|
+ "\n",
|
|
|
+ " # --- Round 3: 智能兜底 ---\n",
|
|
|
+ " # 既然都没完美匹配，就从候选池里挑一个最便宜的，或者直接硬取第一个\n",
|
|
|
+ " \n",
|
|
|
+ " best_fallback = None\n",
|
|
|
+ " status_msg = \"暂无报价\"\n",
|
|
|
+ " \n",
|
|
|
+ " if all_candidates:\n",
|
|
|
+ " # 按价格排序，取最低的那个\n",
|
|
|
+ " best_fallback = sorted(all_candidates, key=lambda x: x['price_val'])[0]\n",
|
|
|
+ " status_msg = f\"约 {best_fallback['price_val']}元 (⚠️超预算)\"\n",
|
|
|
+ " elif results_1:\n",
|
|
|
+ " # 连价格都没搜到，就硬取第一条文本结果\n",
|
|
|
+ " for res in results_1:\n",
|
|
|
+ " if res.get('type') == 'text':\n",
|
|
|
+ " best_fallback = res\n",
|
|
|
+ " break\n",
|
|
|
+ " \n",
|
|
|
+ " if best_fallback:\n",
|
|
|
+ " # 补图逻辑\n",
|
|
|
+ " if not best_fallback.get('img') and fallback_img:\n",
|
|
|
+ " best_fallback['img'] = fallback_img\n",
|
|
|
+ " return best_fallback, status_msg, current_kw\n",
|
|
|
+ " \n",
|
|
|
+ " return None, \"搜索失败\", current_kw\n"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "code",
|
|
|
+ "execution_count": null,
|
|
|
+ "id": "4ec82cc7",
|
|
|
+ "metadata": {},
|
|
|
+ "outputs": [],
|
|
|
+ "source": [
|
|
|
+ "# ==========================================\n",
|
|
|
+ "# 🚀 主执行流程\n",
|
|
|
+ "# ==========================================\n",
|
|
|
+ "\n",
|
|
|
+ "if not user_input_data:\n",
|
|
|
+ " print(\"❌ 未加载用户数据\")\n",
|
|
|
+ "else:\n",
|
|
|
+ " # 0. 解析预算\n",
|
|
|
+ " budget_min, budget_max = parse_budget_range(user_input_data.get('预算', ''))\n",
|
|
|
+ " budget_limit = budget_max * 1.2 # 允许超标 20%\n",
|
|
|
+ " print(f\"\\n💰 预算红线: {budget_max}元 (容忍至 {budget_limit}元)\")\n",
|
|
|
+ "\n",
|
|
|
+ " # 1. 军师制定策略\n",
|
|
|
+ " profile_text = \"\\n\".join([f\"- {k}: {v if v else '未知/不限'}\" for k, v in user_input_data.items()])\n",
|
|
|
+ " print(f\"\\n🚀 任务启动...\\n{'-'*40}\")\n",
|
|
|
+ " print(\"\\n🧠 [1/3] 军师正在制定初步策略...\")\n",
|
|
|
+ " search_strategy = profiler_agent.run(f\"请根据以下用户画像制定搜索策略：\\n\\n{profile_text}\")\n",
|
|
|
+ " print(f\"📝 策略: \\n{search_strategy}\")\n",
|
|
|
+ "\n",
|
|
|
+ " # 2. 准备循环\n",
|
|
|
+ " keywords = [k.strip() for k in search_strategy.replace(\"，\", \",\").replace(\"\\n\", \",\").split(',') if k.strip()]\n",
|
|
|
+ " final_items = []\n",
|
|
|
+ " hunter = BatchSearchTool()\n",
|
|
|
+ "\n",
|
|
|
+ " print(f\"\\n🔄 进入处理流程 (共 {len(keywords)} 个商品)...\")\n",
|
|
|
+ "\n",
|
|
|
+ " for index, kw in enumerate(keywords):\n",
|
|
|
+ " print(f\"\\n 👉 [商品 {index+1}/{len(keywords)}] 正在处理: {kw}\")\n",
|
|
|
+ " \n",
|
|
|
+ " # 调用智能搜索函数\n",
|
|
|
+ " valid_result, price_status, final_kw = find_best_product(hunter, profiler_agent, kw, budget_limit)\n",
|
|
|
+ " \n",
|
|
|
+ " if not valid_result:\n",
|
|
|
+ " print(\" ❌ 彻底无数据，跳过。\")\n",
|
|
|
+ " continue\n",
|
|
|
+ " \n",
|
|
|
+ " # === 生成文案 ===\n",
|
|
|
+ " product_name = valid_result.get('title', final_kw)\n",
|
|
|
+ " \n",
|
|
|
+ " print(f\" ✍️ 正在撰写文案: {product_name[:15]}...\")\n",
|
|
|
+ " pitch_prompt = f\"\"\"\n",
|
|
|
+ " 商品：{product_name}\n",
|
|
|
+ " 价格：{price_status}\n",
|
|
|
+ " 卖点片段：{valid_result.get('content', '')[:200]}...\n",
|
|
|
+ " \n",
|
|
|
+ " 请写一句30字以内的种草文案。\n",
|
|
|
+ " \"\"\"\n",
|
|
|
+ " pitch = pitcher_agent.run(pitch_prompt)\n",
|
|
|
+ " \n",
|
|
|
+ " # 整理最终数据\n",
|
|
|
+ " final_items.append({\n",
|
|
|
+ " \"name\": final_kw, \n",
|
|
|
+ " \"title_full\": product_name,\n",
|
|
|
+ " \"price\": price_status,\n",
|
|
|
+ " \"desc\": pitch.replace(\"\\n\", \" \").strip(),\n",
|
|
|
+ " \"img\": valid_result.get('img', ''),\n",
|
|
|
+ " \"link\": valid_result.get('url', '')\n",
|
|
|
+ " })\n",
|
|
|
+ " print(f\" ✅ 已收录 (状态: {price_status})\")"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "0197ef43",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "\n",
|
|
|
+ "### 第6部分：输出礼物计划"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "code",
|
|
|
+ "execution_count": null,
|
|
|
+ "id": "8426fb13",
|
|
|
+ "metadata": {},
|
|
|
+ "outputs": [],
|
|
|
+ "source": [
|
|
|
+ "# --- 4. 渲染与保存 (保持不变) ---\n",
|
|
|
+ "print(f\"\\n💾 正在生成最终报告...\")\n",
|
|
|
+ "if not final_items:\n",
|
|
|
+ " final_md = \"很抱歉，网络搜索似乎出现了问题，未能获取到任何商品信息。\"\n",
|
|
|
+ "else:\n",
|
|
|
+ " table_header = \"| 🎁 礼物名称 | 💰 价格 | ✨ 种草理由 | 🖼️ 图片/链接 |\\n| :--- | :--- | :--- | :--- |\\n\"\n",
|
|
|
+ " table_rows = []\n",
|
|
|
+ " for item in final_items:\n",
|
|
|
+ " name = item['name'].replace(\"|\", \"/\")\n",
|
|
|
+ " price = item['price']\n",
|
|
|
+ " desc = item['desc'].replace(\"|\", \"/\")\n",
|
|
|
+ " link = item['link']\n",
|
|
|
+ " img = item['img']\n",
|
|
|
+ " \n",
|
|
|
+ " if img and img.startswith(\"http\"):\n",
|
|
|
+ " media = f\"[]({link})\"\n",
|
|
|
+ " else:\n",
|
|
|
+ " media = f\"[点击购买]({link})\"\n",
|
|
|
+ " \n",
|
|
|
+ " table_rows.append(f\"| [{name}]({link}) | {price} | {desc} | {media} |\")\n",
|
|
|
+ " final_md = table_header + \"\\n\".join(table_rows)\n",
|
|
|
+ "filename = \"outputs/gift_plan_output.md\"\n",
|
|
|
+ "with open(filename, \"w\", encoding=\"utf-8\") as f:\n",
|
|
|
+ " f.write(final_md)\n",
|
|
|
+ "print(f\"🎉 任务完成！文件已保存: {os.path.abspath(filename)}\")"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "d8b884b5",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "### 第7部分：总结与展望"
|
|
|
+ ]
|
|
|
+ },
|
|
|
+ {
|
|
|
+ "cell_type": "markdown",
|
|
|
+ "id": "08b0b8e8",
|
|
|
+ "metadata": {},
|
|
|
+ "source": [
|
|
|
+ "#### 实现的功能\n",
|
|
|
+ "- 基于用户输入的个人信息，生成符合预算的礼物建议\n",
|
|
|
+ "- 支持用户自定义预算范围、节日、个人喜好等\n",
|
|
|
+ "- 利用搜索引擎获取最新的商品信息和价格\n",
|
|
|
+ "- 提供可视化的建议结果展示\n",
|
|
|
+ "#### 遇到的挑战\n",
|
|
|
+ "- 如何解决大模型的幻觉问题\n",
|
|
|
+ "- 如何解决上下文过长导致提取失败\n",
|
|
|
+ "- 如何设计合理的提示词，优化大模型的输出\n",
|
|
|
+ "\n",
|
|
|
+ "#### 未来改进方向\n",
|
|
|
+ "- 前端交互：新增前端页面，提供更好的用户交互体验\n",
|
|
|
+ "- 数据源增强：接入百度或BigGo的MCP Server，获取更精准的实时价格和商品信息\n",
|
|
|
+ "- 丰富选项：增加更多的个人喜好选项，如喜欢的商品类型、品牌等\n"
|
|
|
+ ]
|
|
|
+ }
|
|
|
+ ],
|
|
|
+ "metadata": {
|
|
|
+ "kernelspec": {
|
|
|
+ "display_name": "ai_3.10",
|
|
|
+ "language": "python",
|
|
|
+ "name": "python3"
|
|
|
+ },
|
|
|
+ "language_info": {
|
|
|
+ "codemirror_mode": {
|
|
|
+ "name": "ipython",
|
|
|
+ "version": 3
|
|
|
+ },
|
|
|
+ "file_extension": ".py",
|
|
|
+ "mimetype": "text/x-python",
|
|
|
+ "name": "python",
|
|
|
+ "nbconvert_exporter": "python",
|
|
|
+ "pygments_lexer": "ipython3",
|
|
|
+ "version": "3.10.19"
|
|
|
+ }
|
|
|
+ },
|
|
|
+ "nbformat": 4,
|
|
|
+ "nbformat_minor": 5
|
|
|
+}
|
jack6249
