InsightRadar/backend/app/services/fetcher_service.py

# app/services/fetcher_service.py
import os
import hashlib
from datetime import timedelta

import httpx
import json
import numpy as np
from dotenv import load_dotenv
from sklearn.metrics.pairwise import cosine_similarity
from sentence_transformers import SentenceTransformer
from app.database import SessionLocal
from app.models.models import (
    InfoSource, TrendingEvent, NewsArticle, DataSyncTask, TaskStatus,
    HeadlineRevision, RankingLog, SourceType, utcnow, UnifiedEvent
)

# 加载环境变量
load_dotenv()
hf_token = os.getenv("HF_TOKEN")
SIMILARITY_THRESHOLD = float(os.getenv("SIMILARITY_THRESHOLD", 0.72))
API_BASE_URL = os.getenv("API_BASE_URL", "https://newsnow.busiyi.world/api/s")
EMBEDDING_MODEL_PATH = os.getenv("EMBEDDING_MODEL_PATH", "")

print("正在加载 BAAI/bge-m3 向量模型...")
# 全局单例
embedder_model = SentenceTransformer(EMBEDDING_MODEL_PATH, local_files_only=True, device="cuda")
print("模型加载完成。")


def generate_md5(text: str) -> str:
    """生成32位MD5哈希值作为全局唯一指纹"""
    return hashlib.md5(text.encode('utf-8')).hexdigest()


def generate_embeddings_batch(texts: list[str]) -> dict:
    """批量生成向量，返回 {text: (embedding_json, numpy_array)}"""
    if not texts:
        return {}
    
    unique_texts = list(set(texts))
    raw_vecs = embedder_model.encode(unique_texts, normalize_embeddings=True, show_progress_bar=False)
    
    result = {}
    for text, raw_vec in zip(unique_texts, raw_vecs):
        truncated_vec = [round(float(x), 5) for x in raw_vec]
        emb_json = json.dumps(truncated_vec, separators=(',', ':'))
        result[text] = (emb_json, raw_vec)
        
    return result


class UnifiedEventClusterer:
    def __init__(self, db):
        self.db = db
        three_days_ago = utcnow() - timedelta(days=3)
        recent_events = db.query(UnifiedEvent).filter(
            UnifiedEvent.created_at >= three_days_ago
        ).order_by(UnifiedEvent.created_at.desc()).limit(300).all()
        
        self.event_vectors = []
        self.event_ids = []
        for ev in recent_events:
            if ev.center_embedding:
                self.event_vectors.append(np.array(json.loads(ev.center_embedding)))
                self.event_ids.append(ev.id)

    def match_or_create(self, title: str, embedding_json: str, new_vec: np.ndarray) -> int:
        if self.event_vectors:
            # 批量矩阵计算相似度
            sim_scores = cosine_similarity([new_vec], self.event_vectors)[0]
            max_idx = np.argmax(sim_scores)

            if sim_scores[max_idx] >= SIMILARITY_THRESHOLD:
                matched_event_id = self.event_ids[max_idx]
                # 更新热度
                matched_event = self.db.query(UnifiedEvent).get(matched_event_id)
                if matched_event:
                    matched_event.hot_score += 1
                return matched_event_id

        # 没匹配到，创建一个新的统一大事件
        new_unified = UnifiedEvent(
            unified_title=title,
            center_embedding=embedding_json,
            hot_score=1  # 初始热度
        )
        self.db.add(new_unified)
        self.db.flush()  # 获取自增的主键 ID
        
        # 更新缓存
        self.event_vectors.append(new_vec)
        self.event_ids.append(new_unified.id)
        
        return new_unified.id


def process_hot_trend_item(db, source, item, index: int, external_id: str, existing_event, embeddings_dict: dict, clusterer: UnifiedEventClusterer):
    """
    处理【热搜/短新闻】的业务逻辑，现已加入 AI 聚类功能
    """
    title = item.get("title")
    item_url = item.get("url", "")

    event_to_log = None

    # 核心逻辑：查重后再决定是否调用模型
    if existing_event:
        # 场景 A1：老熟人
        if existing_event.current_headline != title:
            # 标题被暗改，此时需要重新算一次 Embedding
            new_embedding_json, _ = embeddings_dict[title]

            revision = HeadlineRevision(
                event_id=existing_event.id,
                previous_headline=existing_event.current_headline,
                revised_headline=title
            )
            db.add(revision)
            existing_event.current_headline = title
            existing_event.title_embedding = new_embedding_json  # 更新为新标题的语义向量
            # 注：这里不改变它所属的 unified_event_id，因为大体还是同一件事

        existing_event.current_ranking = index
        existing_event.event_url = item_url
        event_to_log = existing_event

    else:
        # 场景 A2：这是一条彻底的全新热搜
        # 1. 计算向量
        new_embedding_json, new_vec = embeddings_dict[title]

        # 2. 扔进聚类中枢找归宿
        matched_event_id = clusterer.match_or_create(title, new_embedding_json, new_vec)

        # 3. 落库
        new_event = TrendingEvent(
            source_id=source.id,
            external_id=external_id,
            current_headline=title,
            event_url=item_url,
            current_ranking=index,
            title_embedding=new_embedding_json,  # 存入向量
            unified_event_id=matched_event_id  # 挂载到大事件下
        )
        db.add(new_event)
        db.flush()
        event_to_log = new_event

    # 强制记录排名轨迹
    rank_log = RankingLog(
        event_id=event_to_log.id,
        ranking_position=index
    )
    db.add(rank_log)


def process_rss_feed_item(db, source, item, external_id: str, existing_article):
    """
    处理【长文章/传统订阅】分支的核心业务逻辑 (写入 NewsArticle 表)
    """
    title = item.get("title")
    item_url = item.get("url", "")

    if existing_article:
        # 文章若存在，仅更新基础字段
        existing_article.article_title = title
        existing_article.article_url = item_url
    else:
        # 全新文章入库
        new_article = NewsArticle(
            source_id=source.id,
            external_id=external_id,
            article_title=title,
            article_url=item_url,
        )
        db.add(new_article)


def process_source_data(db, source, items: list) -> int:
    """
    数据清洗与路由分发层：
    遍历 API 返回的 items，生成唯一指纹，并路由到不同的处理模块。
    采用批量查重和批量向量计算优化性能，避免数据库锁死。
    返回成功处理的条目数量。
    """
    saved_count = 0
    platform_id = source.home_url

    # 1. 批量计算外部 ID 并聚合要计算的文本
    valid_items = []
    external_ids = []
    for item in items:
        title = item.get("title")
        if not title:
            continue
        item_url = item.get("url", "")
        raw_id = item.get("id") or item_url or title
        external_id = generate_md5(f"{platform_id}_{raw_id}")
        
        valid_items.append((item, external_id))
        external_ids.append(external_id)
        
    if not valid_items:
        return 0

    # 2. 批量数据库查重
    existing_events_dict = {}
    existing_articles_dict = {}
    
    if source.source_type in (SourceType.HOT_TREND, SourceType.API):
        existing_events = db.query(TrendingEvent).filter(
            TrendingEvent.source_id == source.id,
            TrendingEvent.external_id.in_(external_ids)
        ).all()
        existing_events_dict = {ev.external_id: ev for ev in existing_events}
    elif source.source_type == SourceType.RSS_FEED:
        existing_articles = db.query(NewsArticle).filter(
            NewsArticle.source_id == source.id,
            NewsArticle.external_id.in_(external_ids)
        ).all()
        existing_articles_dict = {art.external_id: art for art in existing_articles}

    # 3. 筛选出需要进行大模型向量运算的文本
    texts_to_embed = []
    if source.source_type in (SourceType.HOT_TREND, SourceType.API):
        for item, external_id in valid_items:
            title = item.get("title")
            existing_event = existing_events_dict.get(external_id)
            if existing_event:
                if existing_event.current_headline != title:
                    texts_to_embed.append(title)
            else:
                texts_to_embed.append(title)
                
    # 4. 批量执行大模型推理
    embeddings_dict = generate_embeddings_batch(texts_to_embed)
    
    # 初始化聚类器（只在热搜模式下需要，且只初始化一次）
    clusterer = None
    if source.source_type in (SourceType.HOT_TREND, SourceType.API):
        clusterer = UnifiedEventClusterer(db)

    # 5. 核心路由分流落库
    for index, (item, external_id) in enumerate(valid_items, 1):
        if source.source_type in (SourceType.HOT_TREND, SourceType.API):
            existing_event = existing_events_dict.get(external_id)
            process_hot_trend_item(
                db, source, item, index, external_id, 
                existing_event, embeddings_dict, clusterer
            )
        elif source.source_type == SourceType.RSS_FEED:
            existing_article = existing_articles_dict.get(external_id)
            process_rss_feed_item(db, source, item, external_id, existing_article)

        saved_count += 1

    return saved_count


async def fetch_and_save_trending_data():
    """
    调度层：负责网络请求、数据库事务管理和异常监控隔离。
    """
    print(f"[{utcnow()}] 开始执行定时抓取任务...")

    # 获取启用的信息源 - 这个只读操作用一个短连接
    with SessionLocal() as db:
        sources = db.query(InfoSource).filter(InfoSource.is_enabled == True).all()
        if not sources:
            print("没有找到启用的信息源，任务结束。")
            return
            
        # 我们把 source 的信息提前提取出来，避免在异步中长期持有 session
        source_configs = [
            {
                "id": s.id,
                "home_url": s.home_url,
                "source_name": s.source_name,
                "source_type": s.source_type
            }
            for s in sources
        ]

    # 伪装请求头，规避反爬
    custom_headers = {
        "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/145.0.0.0 Safari/537.36",
        "Accept": "application/json, text/plain, */*",
        "Referer": "https://newsnow.busiyi.world/",
        "Origin": "https://newsnow.busiyi.world"
    }

    async with httpx.AsyncClient(timeout=15.0, headers=custom_headers) as client:
        for s_config in source_configs:
            platform_id = s_config["home_url"]
            if not platform_id:
                continue

            url = f"{API_BASE_URL}?id={platform_id}&latest"

            try:
                # 1. 网络请求（可能耗时较长，不要包在 db session 里）
                response = await client.get(url)
                response.raise_for_status()
                data_json = response.json()
                items = data_json.get("items", [])
                
                # 2. 数据库事务操作（尽量短，单独使用 session）
                with SessionLocal() as db:
                    # 重新从短 session 中获取 source 实例，以免 detached
                    source = db.query(InfoSource).get(s_config["id"])
                    if not source:
                        continue
                        
                    task_log = DataSyncTask(source_id=source.id, items_fetched=0)
                    try:
                        # 调用数据处理层
                        saved_count = process_source_data(db, source, items)

                        # 业务事务成功提交
                        task_log.items_fetched = saved_count
                        task_log.task_status = TaskStatus.SUCCESS
                        db.add(task_log)
                        db.commit()
                        print(f"[{source.source_name}] ({source.source_type}) 成功抓取并更新了 {saved_count} 条数据")
                    except Exception as e:
                        db.rollback()
                        raise e # 抛出给外层捕获记录日志
                        
            except Exception as e:
                # 异常拦截与错误隔离，另起一个超短事务记录日志
                with SessionLocal() as log_db:
                    try:
                        new_task_log = DataSyncTask(source_id=s_config["id"], items_fetched=0)
                        new_task_log.task_status = TaskStatus.ERROR
                        new_task_log.error_trace = str(e)
                        log_db.add(new_task_log)
                        log_db.commit()
                        print(f"[{s_config['source_name']}] 抓取失败: {e}")
                    except Exception as inner_e:
                        log_db.rollback()
                        print(f"[{s_config['source_name']}] 抓取失败，且日志写入失败: {e}, {inner_e}")