#实战项目三：语义搜索与问答系统

# data_prepare.py
import pandas as pd

# FAQ 数据示例
faq_data = [
    {
        "question": "密码忘记了怎么办？",
        "answer": "请访问登录页面，点击"忘记密码"，输入注册邮箱，系统会发送重置链接到您的邮箱。或者联系客服人工处理。",
        "category": "账户"
    },
    {
        "question": "如何修改个人信息？",
        "answer": "登录后点击右上角头像 → "个人中心" → "编辑资料"，即可修改昵称、头像、简介等信息。",
        "category": "账户"
    },
    {
        "question": "订阅如何取消？",
        "answer": "登录后进入"我的订阅"页面，点击"取消订阅"按钮即可。取消后服务将在当前周期结束后停止。",
        "category": "订阅"
    },
    {
        "question": "支持哪些支付方式？",
        "answer": "我们支持微信支付、支付宝、信用卡（Visa/Mastercard）、PayPal 等多种支付方式。",
        "category": "支付"
    },
    # ... 更多 FAQ 条目
]

df = pd.DataFrame(faq_data)
df.to_json("faq.jsonl", orient="records", lines=True, force_ascii=False)

# vectorize.py
from sentence_transformers import SentenceTransformer
import numpy as np
import faiss

# 加载 Sentence-BERT 模型
model = SentenceTransformer("all-MiniLM-L6-v2")  # 轻量，中文可用 paraphrase-multilingual

# 向量化 FAQ
questions = df["question"].tolist()
question_embeddings = model.encode(questions, show_progress_bar=True)
embeddings = np.array(question_embeddings).astype("float32")

# L2 归一化（余弦相似度等价于内积）
norms = np.linalg.norm(embeddings, axis=1, keepdims=True)
embeddings = embeddings / norms

# 构建 FAISS 索引
d = embeddings.shape[1]  # 向量维度
index = faiss.IndexFlatIP(d)  # 内积索引（需归一化）
index.add(embeddings)

# 保存
faiss.write_index(index, "faq.index")
df.to_pickle("faq.pkl")

print(f"FAQ 条目数：{len(questions)}，向量维度：{d}")

# retrieve.py
from sentence_transformers import SentenceTransformer
import faiss
import pandas as pd

model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2")
index = faiss.read_index("faq.index")
faq_df = pd.read_pickle("faq.pkl")

def semantic_search(query, top_k=3):
    """语义检索"""
    # 编码查询
    query_vec = model.encode([query]).astype("float32")
    query_vec = query_vec / np.linalg.norm(query_vec)

    # 检索
    distances, indices = index.search(query_vec, top_k)

    # 返回结果
    results = []
    for dist, idx in zip(distances[0], indices[0]):
        if idx >= 0:  # 有效结果
            results.append({
                "question": faq_df.iloc[idx]["question"],
                "answer": faq_df.iloc[idx]["answer"],
                "score": float(dist),
            })
    return results

# 示例
results = semantic_search("密码忘了怎么重置")
for r in results:
    print(f"问题：{r['question']}")
    print(f"答案：{r['answer']}")
    print(f"相似度：{r['score']:.4f}\n")

# rag_qa.py
from openai import OpenAI
client = OpenAI()

def rag_answer(question, top_k=3):
    """检索 + 生成（RAG）"""
    # 1. 检索相关 FAQ
    results = semantic_search(question, top_k)

    if not results:
        return {"answer": "抱歉，没有找到相关信息。", "sources": []}

    # 2. 构建上下文
    context = "\n\n".join([
        f"FAQ {i+1}：{r['question']}\n答案：{r['answer']}"
        for i, r in enumerate(results)
    ])

    # 3. 构建 Prompt
    prompt = f"""你是一个智能客服助手。根据以下 FAQ 信息回答用户问题。

FAQ 信息：
{context}

用户问题：{question}

要求：
1. 如果 FAQ 中有相关信息，用自己的话总结回答
2. 如果没有相关信息，礼貌地说明无法解答
3. 回答控制在 100 字以内
4. 回答后注明参考的 FAQ 编号
"""

    # 4. 调用 LLM
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": prompt}],
        max_tokens=300,
    )

    answer = response.choices[0].message.content

    return {
        "answer": answer,
        "sources": [{"question": r["question"], "score": r["score"]} for r in results]
    }

# 示例
result = rag_answer("密码忘记了怎么处理")
print(result["answer"])
print("参考来源：", result["sources"])

# app.py
from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI(title="FAQ 智能问答系统")

class QuestionRequest(BaseModel):
    question: str
    top_k: int = 3

@app.post("/qa")
def qa_endpoint(req: QuestionRequest):
    return rag_answer(req.question, req.top_k)

@app.post("/search")
def search_endpoint(req: QuestionRequest):
    results = semantic_search(req.question, req.top_k)
    return {"results": results}

@app.get("/health")
def health():
    return {"status": "ok"}

# 启动：uvicorn app:app --host 0.0.0.0 --port 8000

# ui.py
import streamlit as st
import requests

st.title("FAQ 智能问答系统")

question = st.text_input("请输入您的问题：", placeholder="例如：如何取消订阅？")

if st.button("提问"):
    if question:
        with st.spinner("思考中..."):
            resp = requests.post(
                "http://localhost:8000/qa",
                json={"question": question}
            )
            result = resp.json()

        st.success(result["answer"])

        with st.expander("查看参考来源"):
            for i, src in enumerate(result["sources"]):
                st.write(f"**{i+1}. {src['question']}** (相似度: {src['score']:.3f})")