{"repo": "bobghugjkl/agentlearning", "files": {"langchain-learning/week1/day05/02_extract.py": {"37bc8a5edf2cb2ab7356d4268fa9dd9cb3b5f52f": {"t": 1786068240.7873046, "n": 1597, "h": "e7f52eb980cbad5d", "c": "import os\nfrom dotenv import load_dotenv\nfrom langchain.chat_models import init_chat_model\nfrom langchain_core.prompts import ChatPromptTemplate\nfrom pydantic import BaseModel, Field\nfrom typing import List\nfrom langchain_core.runnables import RunnableWithFallbacks\nload_dotenv()\n\nmodel = init_chat_model(\n    \"deepseek-chat\",\n    model_provider=\"deepseek\",\n    api_key=os.getenv(\"DEEPSEEK_API_KEY\"),\n    temperature=0,\n)\n\n# 定义要抽取的结构\nclass Person(BaseModel):\n    name: str = Field(description=\"人物姓名\")\n    role: str = Field(description=\"人物身份或职业\")\n    fact: str = Field(description=\"关于这个人的一个关键事实\")\n\nclass ArticleInfo(BaseModel):\n    topic: str = Field(description=\"文章主题，一句话概括\")\n    persons: List[Person] = Field(description=\"文章中提到的人物列表\")\n    key_takeaway: str = Field(description=\"最重要的结论，一句话\")\n\n# 构建 chain\nprompt = ChatPromptTemplate.from_messages([\n    (\"system\", \"你是一个信息抽取助手，从文章中抽取结构化信息。\"),\n    (\"human\", \"请从下面这段文字中抽取信息：\\n\\n{text}\"),\n])\n\nstructured_model = model.with_structured_output(ArticleInfo)\nchain = prompt | structured_model\n\n# 测试文章\narticle = \"\"\"\n乃木坂46成立于2011年，是AKB48的官方竞争对手。\n白石麻衣作为团队的顶尖成员，以超模级颜值著称，\n目前已转型为专业模特和演员。\n斋藤飞鸟是继白石麻衣之后的王牌成员，\n以清冷气质和强大的舞台表现力闻名，\n2023年从团体毕业后开始专注演员事业。\n\"\"\"\n\nresult = chain.invoke({\"text\": article})\n\nprint(f\"主题：{result.topic}\")\nprint(f\"\\n人物：\")\nfor person in result.persons:\n    print(f\"  - {person.name}（{person.role}）：{person.fact}\")\nprint(f\"\\n结论：{result.key_takeaway}\")\n\n# 加重试：失败最多重试2次\n\n\nreliable_chain = chain.with_retry(\n    stop_after_attempt=3,  # 最多试3次\n)\n\nresult2 = reliable_chain.invoke({\"text\": article})\nprint(f\"\\n=== 带重试的结果 ===\")\nprint(f\"主题：{result2.topic}\")"}}}}