JWLLL/main_online.py

# main_online.py
# 搜索推荐算法服务的主入口

import json
import numpy as np
import difflib
from flask import Flask, request, jsonify, Response
import pymysql
import jieba
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import pypinyin
from flask_cors import CORS
import re
import Levenshtein
import os
import logging

# 设置日志
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger("allpt-search")

# 导入Word2Vec辅助模块
try:
    from word2vec_helper import get_word2vec_helper, expand_query, get_similar_words
    WORD2VEC_ENABLED = True
    logger.info("Word2Vec模块已加载")
except ImportError as e:
    logger.warning(f"Word2Vec模块加载失败: {e}，将使用传统搜索")
    WORD2VEC_ENABLED = False

# 数据库配置
DB_CONFIG = {
    "host": "10.126.59.25",
    "port": 3306,
    "user": "root",
    "password": "123456",
    "database": "redbook",
    "charset": "utf8mb4"
}

def get_db_conn():
    return pymysql.connect(**DB_CONFIG)

def get_pinyin(text):
    # 返回字符串的全拼音（不带声调，全部小写），支持英文直接返回
    if not text:
        return ""
    import re
    # 如果全是英文，直接返回小写
    if re.fullmatch(r'[a-zA-Z]+', text):
        return text.lower()
    return ''.join([p[0] for p in pypinyin.pinyin(text, style=pypinyin.NORMAL)])

def get_pinyin_initials(text):
    # 返回字符串的首字母拼音（全部小写），支持英文直接返回
    if not text:
        return ""
    import re
    if re.fullmatch(r'[a-zA-Z]+', text):
        return text.lower()
    return ''.join([p[0][0] for p in pypinyin.pinyin(text, style=pypinyin.NORMAL)])

# 新增词语相似度计算函数
def word_similarity(word1, word2):
    """计算两个词的相似度，支持拼音匹配"""
    # 直接匹配
    if word1 == word2:
        return 1.0
    
    # 拼音匹配
    if get_pinyin(word1) == get_pinyin(word2):
        return 0.9
    
    # 拼音首字母匹配
    if get_pinyin_initials(word1) == get_pinyin_initials(word2):
        return 0.7
    
    # 字符串相似度
    return difflib.SequenceMatcher(None, word1, word2).ratio()

def semantic_title_similarity(query, title):
    """计算查询词与标题的语义相似度"""
    # 分词
    query_words = list(jieba.cut(query))
    title_words = list(jieba.cut(title))
    
    if not query_words or not title_words:
        return 0.0
    
    # 计算每个查询词与标题词的最大相似度
    max_similarities = []
    key_matches = 0  # 关键词精确匹配数量
    
    for q_word in query_words:
        if len(q_word.strip()) <= 1:  # 忽略单字，减少噪音
            continue
            
        word_sims = [word_similarity(q_word, t_word) for t_word in title_words]
        if word_sims:
            max_sim = max(word_sims)
            max_similarities.append(max_sim)
            if max_sim > 0.85:  # 认为是关键词匹配
                key_matches += 1
    
    if not max_similarities:
        return 0.0
    
    # 计算平均相似度
    avg_sim = sum(max_similarities) / len(max_similarities)
    
    # 权重计算: 平均相似度占70%，关键词匹配率占30%
    key_match_ratio = key_matches / len(query_words) if query_words else 0
    
    # 标题中包含完整查询短语时给予额外加分
    exact_bonus = 0.3 if query in title else 0
    
    return 0.7 * avg_sim + 0.3 * key_match_ratio + exact_bonus

# 添加语义关联词典，用于增强搜索能力
def load_semantic_mappings():
    """
    加载语义关联映射表，用于增强搜索语义理解
    返回包含语义映射关系的字典
    """
    # 初始化空字典，所有映射将从配置文件加载
    mappings = {}
    
    # 从配置文件加载映射
    try:
        config_path = os.path.join(os.path.dirname(__file__), "semantic_config.json")
        if os.path.exists(config_path):
            with open(config_path, 'r', encoding='utf-8') as f:
                mappings = json.load(f)
            logger.info(f"已从配置文件加载 {len(mappings)} 个语义映射")
        else:
            logger.warning(f"语义配置文件不存在: {config_path}")
    except Exception as e:
        logger.error(f"加载语义配置文件失败: {e}")
    
    return mappings

# 初始化语义映射
SEMANTIC_MAPPINGS = load_semantic_mappings()

def expand_search_keywords(keyword):
    """
    扩展搜索关键词，增加语义关联词
    """
    expanded = [keyword]
    
    # 分词处理
    words = list(jieba.cut(keyword))
    logger.info(f"关键词 '{keyword}' 分词结果: {words}")  # 记录分词结果
    
    # 分别对每个分词进行语义扩展
    for word in words:
        if word in SEMANTIC_MAPPINGS:
            # 添加语义关联词
            mapped_words = SEMANTIC_MAPPINGS[word]
            expanded.extend(mapped_words)
            logger.info(f"语义映射: '{word}' -> {mapped_words}")
            
            # 移除所有特殊处理部分
            # 不再对任何特定关键词如"越狱"进行特殊处理
    
    # Word2Vec扩展 - 如果可用，对分词结果进行Word2Vec扩展
    if WORD2VEC_ENABLED:
        try:
            # 使用单独的变量记录原始扩展结果，方便记录日志
            original_expanded = set(expanded)
            
            # 首先尝试对整个关键词进行扩展
            w2v_expanded = set()
            similar_words = get_similar_words(keyword, topn=3, min_similarity=0.6)
            w2v_expanded.update(similar_words)
            
            # 然后对较长的分词进行扩展
            for word in words:
                if len(word) > 1:  # 忽略单字
                    similar_words = get_similar_words(word, topn=2, min_similarity=0.65)
                    w2v_expanded.update(similar_words)
            
            # 合并结果
            expanded.extend(w2v_expanded)
            
            # 记录日志
            if w2v_expanded:
                logger.info(f"Word2Vec扩展: {keyword} -> {list(w2v_expanded)}")
        except Exception as e:
            # 出错时记录但不中断搜索流程
            logger.error(f"Word2Vec扩展失败: {e}")
            logger.info("将仅使用配置文件中的语义映射")
    
    # 去重
    return list(set(expanded))

# 替换原有的calculate_keyword_relevance函数，采用更通用的相关性算法
def calculate_keyword_relevance(keyword, item):
    """计算搜索关键词与条目的相关性得分"""
    title = item.get('title', '')
    description = item.get('description', '') or ''
    tags = item.get('tags', '') or ''
    category = item.get('category', '') or ''  # 添加category字段
    
    # 初始化得分
    score = 0
    
    # 1. 精确匹配（最高优先级）
    if keyword.lower() == title.lower():
        return 15.0  # 完全匹配给予最高分
    
    # 2. 标题中精确词匹配
    title_words = re.findall(r'\b\w+\b', title.lower())
    if keyword.lower() in title_words:
        score += 10.0  # 作为独立词完全匹配
    
    # 3. 标题包含关键词（部分匹配）
    elif keyword.lower() in title.lower():
        # 计算关键词所占标题比例
        match_ratio = len(keyword) / len(title)
        if match_ratio > 0.5:  # 关键词占标题很大比例
            score += 8.0
        else:
            score += 5.0
    
    # 4. 标题分词匹配
    keyword_words = list(jieba.cut(keyword))
    title_jieba_words = list(jieba.cut(title))
    
    matched_words = 0
    for k_word in keyword_words:
        if len(k_word) > 1:  # 忽略单字
            if k_word in title_jieba_words:
                matched_words += 1
            else:
                # 拼音匹配
                k_pinyin = get_pinyin(k_word)
                for t_word in title_jieba_words:
                    if get_pinyin(t_word) == k_pinyin:
                        matched_words += 0.8
                        break
    
    if len(keyword_words) > 0:
        word_match_ratio = matched_words / len(keyword_words)
        score += 3.0 * word_match_ratio
    
    # 5. 拼音相似度
    keyword_pinyin = get_pinyin(keyword)
    title_pinyin = get_pinyin(title)
    
    if keyword_pinyin == title_pinyin:
        score += 3.5
    elif keyword_pinyin in title_pinyin:
        # 计算拼音在标题中的位置影响
        pos = title_pinyin.find(keyword_pinyin)
        if pos == 0:  # 出现在开头
            score += 3.0
        else:
            score += 2.0
    
    # 6. 编辑距离相似度
    try:
        edit_distance = Levenshtein.distance(keyword.lower(), title.lower())
        max_len = max(len(keyword), len(title))
        if max_len > 0:
            similarity = 1 - (edit_distance / max_len)
            if similarity > 0.7:
                score += 1.5 * similarity
    except:
        similarity = difflib.SequenceMatcher(None, keyword.lower(), title.lower()).ratio()
        if similarity > 0.7:
            score += 1.5 * similarity
    
    # 7. 中文字符重叠检测 - 修改为仅当重叠2个以上汉字或占比超过40%时才计分
    if re.search(r'[\u4e00-\u9fff]', keyword) and re.search(r'[\u4e00-\u9fff]', title):
        cn_chars_keyword = set(re.findall(r'[\u4e00-\u9fff]', keyword))
        cn_chars_title = set(re.findall(r'[\u4e00-\u9fff]', title))
        
        # 计算重叠的汉字集合
        overlapped_chars = cn_chars_keyword & cn_chars_title
        
        # 仅当重叠汉字数量大于1且占比超过阈值时才计分
        if len(overlapped_chars) > 1 and len(cn_chars_keyword) > 0:
            overlap_ratio = len(overlapped_chars) / len(cn_chars_keyword)
            # 增加重叠比例的阈值要求，防止单个汉字导致的误匹配
            if overlap_ratio >= 0.4 or len(overlapped_chars) >= 3:
                score += 2.0 * overlap_ratio
            # 对于非常低的重叠度，不加分，避免无关内容干扰
        
        # 记录日志，帮助调试特定案例
        if keyword == "明日方舟" and "白日梦想家" in title:
            logger.info(f"'明日方舟'与'{title}'的汉字重叠: {overlapped_chars}, 重叠比例: {len(overlapped_chars)/len(cn_chars_keyword) if cn_chars_keyword else 0}")
    
    # 8. 序列资源检测（如"功夫熊猫2"是"功夫熊猫"的系列）
    base_title_match = re.match(r'(.*?)([0-9]+|[一二三四五六七八九十]|：|\:|\s+[0-9]+)', title)
    if base_title_match:
        base_title = base_title_match.group(1).strip()
        if keyword.lower() == base_title.lower():
            score += 2.0
    
    # 9. 标签和描述匹配（增加权重）
    if tags:
        tags_list = tags.split(',')
        if keyword in tags_list:
            score += 1.5  # 提高标签匹配的权重
        elif any(keyword.lower() in tag.lower() for tag in tags_list):
            score += 1.0  # 提高部分匹配的权重
    
    # 描述匹配增强
    if keyword.lower() in description.lower():
        score += 1.5  # 提高描述匹配的权重
        
        # 检查关键词在描述中的位置和上下文
        pos = description.lower().find(keyword.lower())
        if pos >= 0 and pos < len(description) / 3:
            # 关键词出现在描述前1/3部分，可能更重要
            score += 0.5
    
    # 考虑分词匹配描述
    keyword_words = list(jieba.cut(keyword))
    description_words = list(jieba.cut(description))
    matched_desc_words = 0
    for k_word in keyword_words:
        if len(k_word) > 1 and k_word in description_words:
            matched_desc_words += 1
    
    if len(keyword_words) > 0:
        desc_match_ratio = matched_desc_words / len(keyword_words)
        score += 1.0 * desc_match_ratio
    
    # 分类匹配
    if keyword.lower() in category.lower():
        score += 1.0
    
    # 添加语义关联匹配得分
    # 扩展关键词进行匹配
    expanded_keywords = expand_search_keywords(keyword)
    # 检测标题是否包含语义相关词
    for exp_keyword in expanded_keywords:
        if exp_keyword != keyword and exp_keyword in title:  # 避免重复计算原关键词
            score += 1.5  # 一般语义关联
    
    return score

# 创建Flask应用
app = Flask(__name__)
CORS(app)  # 允许所有跨域请求

# 添加init_word2vec函数
def init_word2vec():
    """初始化Word2Vec模型"""
    try:
        helper = get_word2vec_helper()
        if helper.initialized:
            logger.info(f"Word2Vec模型已成功加载，词汇量: {len(helper.model.index_to_key)}, 向量维度: {helper.model.vector_size}")
        else:
            if helper.load_model():
                logger.info(f"Word2Vec模型加载成功，词汇量: {len(helper.model.index_to_key)}, 向量维度: {helper.model.vector_size}")
            else:
                logger.error("Word2Vec模型加载失败")
    except Exception as e:
        logger.error(f"初始化Word2Vec出错: {e}")

# 新的初始化方式:
def initialize_app():
    """应用初始化函数，替代before_first_request装饰器"""
    # 修正：使用正确的函数名
    # 原代码: init_semantic_mapping()
    # 修正为使用已定义的函数名
    global SEMANTIC_MAPPINGS
    SEMANTIC_MAPPINGS = load_semantic_mappings()  # 更新全局语义映射变量
    
    if WORD2VEC_ENABLED:
        init_word2vec()  # 现在这个函数已经定义了

# 在启动应用之前调用初始化函数
initialize_app()

# 测试路由
@app.route('/test', methods=['GET'])
def test():
    import datetime
    return jsonify({"message": "服务器正常运行", "timestamp": str(datetime.datetime.now())})

# 获取单个帖子详情的API
@app.route('/post/<int:post_id>', methods=['GET'])
def get_post_detail(post_id):
    """
    获取单个帖子详情
    """
    logger.info(f"接收到获取帖子详情请求，post_id: {post_id}")
    conn = get_db_conn()
    try:
        with conn.cursor(pymysql.cursors.DictCursor) as cursor:
            # 查询帖子详情，先用简单查询调试
            query = """
            SELECT 
                p.id,
                p.title,
                p.content,
                p.heat,
                p.created_at as create_time,
                p.updated_at as last_active,
                p.status
            FROM posts p 
            WHERE p.id = %s
            """
            logger.info(f"执行查询: {query} with post_id: {post_id}")
            cursor.execute(query, (post_id,))
            post = cursor.fetchone()
            
            logger.info(f"查询结果: {post}")
            
            if not post:
                logger.warning(f"帖子不存在，post_id: {post_id}")
                return jsonify({"error": "帖子不存在"}), 404
            
            # 设置默认值
            post['tags'] = []
            post['category'] = '未分类'
            post['author'] = '匿名用户'
            
            # 格式化时间
            if post['create_time']:
                post['create_time'] = post['create_time'].strftime('%Y-%m-%d %H:%M:%S')
            if post['last_active']:
                post['last_active'] = post['last_active'].strftime('%Y-%m-%d %H:%M:%S')
            
            logger.info(f"返回帖子详情: {post}")
            return Response(json.dumps(post, ensure_ascii=False), mimetype='application/json; charset=utf-8')
    except Exception as e:
        logger.error(f"获取帖子详情失败: {e}")
        import traceback
        traceback.print_exc()
        return jsonify({"error": "服务器内部错误"}), 500
    finally:
        conn.close()

# 搜索功能的API
@app.route('/search', methods=['POST'])
def search():
    """
    搜索功能API
    请求格式：{
        "keyword": "关键词",
        "sort_by": "downloads" | "downloads_asc" | "newest" | "oldest" | "similarity" | "title_asc" | "title_desc",
        "category": "可选，分类名",
        "search_mode": "title" | "title_desc" | "tags" | "all"  # 可选，默认"title",
        "tags": ["标签1", "标签2"]  # 可选，支持传递多个标签
    }
    """
    if request.content_type != 'application/json':
        return jsonify({"error": "Content-Type must be application/json"}), 415

    data = request.get_json()
    keyword = data.get("keyword", "").strip()
    sort_by = data.get("sort_by", "similarity")  # 默认按相似度排序
    category = data.get("category", None)
    search_mode = data.get("search_mode", "title")
    tags = data.get("tags", None)  # 支持传递多个标签

    # 校验参数 - 不管什么模式都要求关键词
    if not (1 <= len(keyword) <= 20):
        return jsonify({"error": "请输入1-20个字符"}), 400

    # 第一阶段：数据库查询获取候选集
    results = []
    conn = get_db_conn()
    try:
        with conn.cursor(pymysql.cursors.DictCursor) as cursor:
            # 首先尝试查询完全匹配的结果
            exact_query = f"""
                SELECT id, title, topic_id, heat, created_at, content
                FROM posts
                WHERE title = %s
            """
            cursor.execute(exact_query, (keyword,))
            exact_matches = cursor.fetchall() or []  # 确保返回列表而非元组
            
            # 扩展关键词，增加语义关联词
            expanded_keywords = expand_search_keywords(keyword)
            logger.info(f"扩展后的关键词: {expanded_keywords}")  # 调试信息
            
            # 构建查询条件
            conditions = []
            params = []
            
            # 标题匹配 - 所有搜索模式都匹配title
            conditions.append("title LIKE %s")
            params.append(f"%{keyword}%")
            
            # 为扩展关键词添加标题匹配条件
            for exp_keyword in expanded_keywords:
                if exp_keyword != keyword:  # 避免重复原关键词
                    conditions.append("title LIKE %s")
                    params.append(f"%{exp_keyword}%")
            
            # 描述匹配
            if search_mode in ["title_desc", "all"]:
                # 原始关键词匹配描述
                conditions.append("content LIKE %s")
                params.append(f"%{keyword}%")
                
                # 扩展关键词匹配描述
                for exp_keyword in expanded_keywords:
                    if exp_keyword != keyword:
                        conditions.append("content LIKE %s")
                        params.append(f"%{exp_keyword}%")
            
            # 标签匹配
            # 暂不处理，后续join实现
            
            # 分类匹配 - 仅在all模式下
            if search_mode == "all":
                # 原始关键词匹配分类
                conditions.append("topic_id LIKE %s")
                params.append(f"%{keyword}%")
                
                # 扩展关键词匹配分类
                for exp_keyword in expanded_keywords:
                    if exp_keyword != keyword:
                        conditions.append("topic_id LIKE %s")
                        params.append(f"%{exp_keyword}%")
            
            # 构建SQL查询
            if conditions:
                where_clause = " OR ".join(conditions)
                logger.info(f"搜索条件: {where_clause}")
                logger.info(f"参数列表: {params}")
                
                if category:
                    where_clause = f"({where_clause}) AND topic_id=%s"
                    params.append(category)
                
                sql = f"""
                    SELECT p.id, p.title, tp.name as category, p.heat, p.created_at, p.content,
                        GROUP_CONCAT(t.name) as tags
                    FROM posts p
                    LEFT JOIN post_tags pt ON p.id = pt.post_id
                    LEFT JOIN tags t ON pt.tag_id = t.id
                    LEFT JOIN topics tp ON p.topic_id = tp.id
                    WHERE {where_clause}
                    GROUP BY p.id
                    LIMIT 500
                """
                
                cursor.execute(sql, params)
                expanded_results = cursor.fetchall()
                logger.info(f"数据库返回记录数: {len(expanded_results) if expanded_results else 0}")
            else:
                expanded_results = []

            # 如果扩展查询和精确匹配都没有结果，获取全部记录进行相关性计算
            if not expanded_results and not exact_matches:
                sql = "SELECT p.id, p.title, tp.name as category, p.heat, p.created_at, p.content, GROUP_CONCAT(t.name) as tags FROM posts p LEFT JOIN post_tags pt ON p.id = pt.post_id LEFT JOIN tags t ON pt.tag_id = t.id LEFT JOIN topics tp ON p.topic_id = tp.id"
                if category:
                    sql += " WHERE p.topic_id=%s"
                    category_params = [category]
                    cursor.execute(sql + " GROUP BY p.id", category_params)
                else:
                    cursor.execute(sql + " GROUP BY p.id")
                
                all_results = cursor.fetchall() or []  # 确保返回列表
            else:
                if isinstance(exact_matches, tuple):
                    exact_matches = list(exact_matches)
                if isinstance(expanded_results, tuple):
                    expanded_results = list(expanded_results)
                all_results = expanded_results + exact_matches
            
            # 对所有结果使用相关性计算规则
            scored_results = []
            for item in all_results:
                # 计算相关性得分
                relevance_score = calculate_keyword_relevance(keyword, item)
                
                # 降低相关性阈值，确保更多结果被保留 (从0.5改为0.1)
                if relevance_score > 0.1:
                    item['relevance_score'] = relevance_score
                    scored_results.append(item)
                    logger.info(f"匹配项: {item['title']}, 相关性得分: {relevance_score}")
            
            # 按相关性得分排序
            scored_results.sort(key=lambda x: x.get('relevance_score', 0), reverse=True)
            
            # 确保精确匹配的结果置顶
            if exact_matches:
                for exact_match in exact_matches:
                    exact_match['relevance_score'] = 20.0  # 超高分确保置顶
                
                # 移除scored_results中已经存在于exact_matches的项
                exact_ids = {item['id'] for item in exact_matches}
                scored_results = [item for item in scored_results if item['id'] not in exact_ids]
                
                # 合并两个结果集
                results = exact_matches + scored_results
            else:
                results = scored_results
            
            # 限制返回结果数量
            results = results[:50]
            
    except Exception as e:
        logger.error(f"搜索出错: {e}")
        import traceback
        traceback.print_exc()
        return jsonify({"error": "搜索系统异常，请稍后再试"}), 500
    finally:
        conn.close()
    
    # 第二阶段：根据指定方式排序
    if results:
        if sort_by == "similarity" or not sort_by:
            # 保持按相关性得分排序，已经排好了
            pass
        elif sort_by == "downloads":
            results.sort(key=lambda x: x.get("download_count", 0), reverse=True)
        elif sort_by == "downloads_asc":
            results.sort(key=lambda x: x.get("download_count", 0))
        elif sort_by == "newest":
            results.sort(key=lambda x: x.get("create_time", ""), reverse=True)
        elif sort_by == "oldest":
            results.sort(key=lambda x: x.get("create_time", ""))
        elif sort_by == "title_asc":
            results.sort(key=lambda x: x.get("title", ""))
        elif sort_by == "title_desc":
            results.sort(key=lambda x: x.get("title", ""), reverse=True)
    
    # 最终处理：清理不需要返回的字段，并将 datetime 转为字符串
    for item in results:
        item.pop("description", None)
        item.pop("tags", None)
        item.pop("relevance_score", None)
        for k, v in item.items():
            if hasattr(v, 'isoformat'):
                item[k] = v.isoformat(sep=' ', timespec='seconds')

    return Response(json.dumps({"results": results}, ensure_ascii=False), mimetype='application/json; charset=utf-8')

# 推荐功能的API
@app.route('/recommend_tags', methods=['POST'])
def recommend_tags():
    """
    推荐功能API
    请求格式：{
        "user_id": "user1",
        "tags": ["标签1", "标签2"]  # 可为空
    }
    """
    if request.content_type != 'application/json':
        return jsonify({"error": "Content-Type must be application/json"}), 415

    data = request.get_json()
    user_id = data.get("user_id")
    tags = set(data.get("tags", []))

    # 查询用户已保存的兴趣标签
    user_tags = set()
    if user_id:
        conn = get_db_conn()
        try:
            with conn.cursor() as cursor:
                cursor.execute("SELECT t.name FROM user_tags ut JOIN tags t ON ut.tag_id = t.id WHERE ut.user_id=%s", (user_id,))
                user_tags = set(row[0] for row in cursor.fetchall())
        finally:
            conn.close()

    # 合并前端传递的tags和用户兴趣标签
    all_tags = list(tags | user_tags)

    if not all_tags:
        return Response(json.dumps({"error": "暂无推荐结果"}, ensure_ascii=False), mimetype='application/json; charset=utf-8'), 200

    conn = get_db_conn()
    try:
        with conn.cursor(pymysql.cursors.DictCursor) as cursor:
            # 优先用tags字段匹配
            # 先查找所有tag_id
            tag_ids = []
            for tag in all_tags:
                cursor.execute("SELECT id FROM tags WHERE name=%s", (tag,))
                row = cursor.fetchone()
                if row:
                    tag_ids.append(row['id'])
            if not tag_ids:
                return Response(json.dumps({"error": "暂无推荐结果"}, ensure_ascii=False), mimetype='application/json; charset=utf-8'), 200
            tag_placeholders = ','.join(['%s'] * len(tag_ids))
            sql = f"""
                SELECT p.id, p.title, tp.name as category, p.heat,
                       GROUP_CONCAT(tg.name) as tags
                FROM posts p
                LEFT JOIN post_tags pt ON p.id = pt.post_id
                LEFT JOIN tags tg ON pt.tag_id = tg.id
                LEFT JOIN topics tp ON p.topic_id = tp.id
                WHERE pt.tag_id IN ({tag_placeholders})
                GROUP BY p.id
                LIMIT 50
            """
            cursor.execute(sql, tuple(tag_ids))
            results = cursor.fetchall()
            # 若无结果，回退title/content模糊匹配
            if not results:
                or_conditions = []
                params = []
                for tag in all_tags:
                    or_conditions.append("p.title LIKE %s OR p.content LIKE %s")
                    params.extend(['%' + tag + '%', '%' + tag + '%'])
                where_clause = ' OR '.join(or_conditions)
                sql = f"""
                    SELECT p.id, p.title, tp.name as category, p.heat,
                           GROUP_CONCAT(tg.name) as tags
                    FROM posts p
                    LEFT JOIN post_tags pt ON p.id = pt.post_id
                    LEFT JOIN tags tg ON pt.tag_id = tg.id
                    LEFT JOIN topics tp ON p.topic_id = tp.id
                    WHERE {where_clause}
                    GROUP BY p.id
                    LIMIT 50
                """
                cursor.execute(sql, tuple(params))
                results = cursor.fetchall()
    finally:
        conn.close()

    if not results:
        return Response(json.dumps({"error": "暂无推荐结果"}, ensure_ascii=False), mimetype='application/json; charset=utf-8'), 200

    return Response(json.dumps({"recommendations": results}, ensure_ascii=False), mimetype='application/json; charset=utf-8')

# 用户兴趣标签管理API（可选）
@app.route('/tags', methods=['POST', 'GET', 'DELETE'])
def user_tags():
    """
    POST: 添加用户兴趣标签
    GET: 查询用户兴趣标签
    DELETE: 删除用户兴趣标签
    """
    if request.method == 'POST':
        if request.content_type != 'application/json':
            return jsonify({"error": "Content-Type must be application/json"}), 415
        data = request.get_json()
        user_id = data.get("user_id")
        tags = data.get("tags", [])
        
        if not user_id:
            return jsonify({"error": "用户ID不能为空"}), 400
        
        # 确保标签列表格式正确
        if isinstance(tags, str):
            tags = [tag.strip() for tag in tags.split(',') if tag.strip()]
        
        if not tags:
            return jsonify({"error": "标签不能为空"}), 400
        
        conn = get_db_conn()
        try:
            with conn.cursor() as cursor:
                # 添加用户标签
                for tag in tags:
                    # 先查找tag_id
                    cursor.execute("SELECT id FROM tags WHERE name=%s", (tag,))
                    tag_row = cursor.fetchone()
                    if tag_row:
                        tag_id = tag_row[0]
                        cursor.execute("REPLACE INTO user_tags (user_id, tag_id) VALUES (%s, %s)", (user_id, tag_id))
                conn.commit()
                # 返回更新后的标签列表
                cursor.execute("SELECT t.name FROM user_tags ut JOIN tags t ON ut.tag_id = t.id WHERE ut.user_id=%s", (user_id,))
                updated_tags = [row[0] for row in cursor.fetchall()]
        finally:
            conn.close()
        return Response(json.dumps({"msg": "添加成功", "tags": updated_tags}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
    elif request.method == 'DELETE':
        if request.content_type != 'application/json':
            return jsonify({"error": "Content-Type must be application/json"}), 415
        data = request.get_json()
        user_id = data.get("user_id")
        tags = data.get("tags", [])
        if not user_id:
            return jsonify({"error": "用户ID不能为空"}), 400
        if not tags:
            return jsonify({"error": "标签不能为空"}), 400
        
        conn = get_db_conn()
        try:
            with conn.cursor() as cursor:
                for tag in tags:
                    cursor.execute("SELECT id FROM tags WHERE name=%s", (tag,))
                    tag_row = cursor.fetchone()
                    if tag_row:
                        tag_id = tag_row[0]
                        cursor.execute("DELETE FROM user_tags WHERE user_id=%s AND tag_id=%s", (user_id, tag_id))
                conn.commit()
                cursor.execute("SELECT t.name FROM user_tags ut JOIN tags t ON ut.tag_id = t.id WHERE ut.user_id=%s", (user_id,))
                remaining_tags = [row[0] for row in cursor.fetchall()]
        finally:
            conn.close()
        return Response(json.dumps({"msg": "删除成功", "tags": remaining_tags}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
    else:  # GET 请求
        user_id = request.args.get("user_id")
        if not user_id:
            return jsonify({"error": "用户ID不能为空"}), 400
        conn = get_db_conn()
        try:
            with conn.cursor() as cursor:
                cursor.execute("SELECT t.name FROM user_tags ut JOIN tags t ON ut.tag_id = t.id WHERE ut.user_id=%s", (user_id,))
                tags = [row[0] for row in cursor.fetchall()]
        finally:
            conn.close()
        return Response(json.dumps({"tags": tags}, ensure_ascii=False), mimetype='application/json; charset=utf-8')

# 添加/user_tags路由作为/tags的别名
@app.route('/user_tags', methods=['POST', 'GET', 'DELETE'])
def user_tags_alias():
    """
    /user_tags路由 - 作为/tags路由的别名
    POST: 添加用户兴趣标签
    GET: 查询用户兴趣标签
    DELETE: 删除用户兴趣标签
    """
    return user_tags()

# 基于用户的协同过滤推荐API
@app.route('/user_based_recommend', methods=['POST'])
def user_based_recommend():
    """
    基于用户的协同过滤推荐API
    请求格式：{
        "user_id": "user1",
        "top_n": 5
    }
    """
    if request.content_type != 'application/json':
        return jsonify({"error": "Content-Type must be application/json"}), 415

    data = request.get_json()
    user_id = data.get("user_id")
    top_n = int(data.get("top_n", 5))

    if not user_id:
        return jsonify({"error": "用户ID不能为空"}), 400
    
    conn = get_db_conn()
    try:
        with conn.cursor(pymysql.cursors.DictCursor) as cursor:
            # 1. 检查用户是否存在下载记录（收藏或浏览）
            cursor.execute("""
                SELECT COUNT(*) as count
                FROM behaviors
                WHERE user_id = %s AND type IN ('favorite', 'view')
            """, (user_id,))
            result = cursor.fetchone()
            user_download_count = result['count'] if result else 0
            
            logger.info(f"用户 {user_id} 下载记录数: {user_download_count}")
            
            # 如果用户没有足够的行为数据，返回基于热度的推荐
            if user_download_count < 3:
                logger.info(f"用户 {user_id} 下载记录不足，返回热门推荐")
                cursor.execute("""
                    SELECT p.id, p.title, tp.name as category, p.heat
                    FROM posts p
                    LEFT JOIN topics tp ON p.topic_id = tp.id
                    ORDER BY p.heat DESC
                    LIMIT %s
                """, (top_n,))
                popular_seeds = cursor.fetchall()
                return Response(json.dumps({"recommendations": popular_seeds, "type": "popular"}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
            
            # 2. 获取用户已下载（收藏/浏览）的帖子
            cursor.execute("""
                SELECT post_id
                FROM behaviors
                WHERE user_id = %s AND type IN ('favorite', 'view')
            """, (user_id,))
            user_seeds = set(row['post_id'] for row in cursor.fetchall())
            logger.info(f"用户 {user_id} 已下载种子: {user_seeds}")
            
            # 3. 获取所有用户-帖子下载（收藏/浏览）矩阵
            cursor.execute("""
                SELECT user_id, post_id
                FROM behaviors
                WHERE created_at > DATE_SUB(NOW(), INTERVAL 3 MONTH)
                AND user_id <> %s AND type IN ('favorite', 'view')
            """, (user_id,))
            download_records = cursor.fetchall()
            
            if not download_records:
                logger.info(f"没有其他用户的下载记录，返回热门推荐")
                cursor.execute("""
                    SELECT p.id, p.title, tp.name as category, p.heat
                    FROM posts p
                    LEFT JOIN topics tp ON p.topic_id = tp.id
                    ORDER BY p.heat DESC
                    LIMIT %s
                """, (top_n,))
                popular_seeds = cursor.fetchall()
                return Response(json.dumps({"recommendations": popular_seeds, "type": "popular"}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
            
            # 构建用户-物品矩阵
            user_item_matrix = {}
            for record in download_records:
                uid = record['user_id']
                sid = record['post_id']
                if uid not in user_item_matrix:
                    user_item_matrix[uid] = set()
                user_item_matrix[uid].add(sid)
            
            # 4. 计算用户相似度
            similar_users = []
            for other_id, other_seeds in user_item_matrix.items():
                if other_id == user_id:
                    continue
                intersection = len(user_seeds.intersection(other_seeds))
                union = len(user_seeds.union(other_seeds))
                if union > 0 and intersection > 0:
                    similarity = intersection / union
                    similar_users.append((other_id, similarity, other_seeds))
            logger.info(f"找到 {len(similar_users)} 个相似用户")
            similar_users.sort(key=lambda x: x[1], reverse=True)
            similar_users = similar_users[:5]
            # 5. 基于相似用户推荐帖子
            candidate_seeds = {}
            for similar_user, similarity, seeds in similar_users:
                logger.info(f"相似用户 {similar_user}, 相似度 {similarity}")
                for post_id in seeds:
                    if post_id not in user_seeds:
                        if post_id not in candidate_seeds:
                            candidate_seeds[post_id] = 0
                        candidate_seeds[post_id] += similarity
            if not candidate_seeds:
                logger.info(f"没有找到候选种子，返回热门推荐")
                cursor.execute("""
                    SELECT p.id, p.title, tp.name as category, p.heat
                    FROM posts p
                    LEFT JOIN topics tp ON p.topic_id = tp.id
                    ORDER BY p.heat DESC
                    LIMIT %s
                """, (top_n,))
                popular_seeds = cursor.fetchall()
                return Response(json.dumps({"recommendations": popular_seeds, "type": "popular"}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
            # 6. 获取推荐帖子的详细信息
            recommended_seeds = sorted(candidate_seeds.items(), key=lambda x: x[1], reverse=True)[:top_n]
            post_ids = [post_id for post_id, _ in recommended_seeds]
            format_strings = ','.join(['%s'] * len(post_ids))
            cursor.execute(f"""
                SELECT p.id, p.title, tp.name as category, p.heat
                FROM posts p
                LEFT JOIN topics tp ON p.topic_id = tp.id
                WHERE p.id IN ({format_strings})
            """, tuple(post_ids))
            result_seeds = cursor.fetchall()
            seed_score_map = {post_id: score for post_id, score in recommended_seeds}
            result_seeds.sort(key=lambda x: seed_score_map.get(x['id'], 0), reverse=True)
            logger.info(f"返回 {len(result_seeds)} 个基于协同过滤的推荐")
            return Response(json.dumps({"recommendations": result_seeds, "type": "collaborative"}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
    except Exception as e:
        logger.error(f"推荐系统错误: {e}")
        import traceback
        traceback.print_exc()
        return Response(json.dumps({"error": "推荐系统异常，请稍后再试", "details": str(e)}, ensure_ascii=False), mimetype='application/json; charset=utf-8')
    finally:
        conn.close()
@app.route('/word2vec_status', methods=['GET'])
def word2vec_status():
    """
    检查Word2Vec模型状态
    返回模型是否加载、词汇量等信息
    """
    if not WORD2VEC_ENABLED:
        return Response(json.dumps({
            "enabled": False,
            "message": "Word2Vec功能未启用"
        }, ensure_ascii=False), mimetype='application/json; charset=utf-8')
    try:
        helper = get_word2vec_helper()
        status = {
            "enabled": WORD2VEC_ENABLED,
            "initialized": helper.initialized,
            "vocab_size": len(helper.model.index_to_key) if helper.model else 0,
            "vector_size": helper.model.vector_size if helper.model else 0
        }
        
        # 测试几个常用词的相似词，展示模型效果
        test_results = {}
        test_words = ["电影", "动作", "科幻", "动漫", "游戏"]
        for word in test_words:
            similar_words = helper.get_similar_words(word, topn=5)
            test_results[word] = similar_words
        
        status["test_results"] = test_results
        return Response(json.dumps(status, ensure_ascii=False), mimetype='application/json; charset=utf-8')
    except Exception as e:
        return Response(json.dumps({
            "enabled": WORD2VEC_ENABLED,
            "initialized": False,
            "error": str(e)
        }, ensure_ascii=False), mimetype='application/json; charset=utf-8')

# 添加一个临时诊断端点
@app.route('/debug_search', methods=['POST'])
def debug_search():
    """临时的调试端点，用于检查数据库中的记录"""
    if request.content_type != 'application/json':
        return jsonify({"error": "Content-Type must be application/json"}), 415

    data = request.get_json()
    keyword = data.get("keyword", "").strip()
    
    conn = get_db_conn()
    try:
        with conn.cursor(pymysql.cursors.DictCursor) as cursor:
            # 尝试查询包含特定词的所有记录
            queries = [
                ("标题中包含关键词", f"SELECT seed_id, title, description, tags FROM pt_seed WHERE title LIKE '%{keyword}%' LIMIT 10"),
                ("描述中包含关键词", f"SELECT seed_id, title, description, tags FROM pt_seed WHERE description LIKE '%{keyword}%' LIMIT 10"),
                ("标签中包含关键词", f"SELECT seed_id, title, description, tags FROM pt_seed WHERE FIND_IN_SET('{keyword}', tags) LIMIT 10"),
                ("肖申克的救赎", "SELECT seed_id, title, description, tags FROM pt_seed WHERE title = '肖申克的救赎'")
            ]
            
            results = {}
            for query_name, query in queries:
                cursor.execute(query)
                results[query_name] = cursor.fetchall()
                
            return Response(json.dumps(results, ensure_ascii=False), mimetype='application/json; charset=utf-8')
    finally:
        conn.close()

"""
接口本地测试方法（可直接运行main_online.py后用curl或Postman测试）：

1. 搜索接口
curl -X POST http://127.0.0.1:5000/search -H "Content-Type: application/json" -d '{"keyword":"电影","sort_by":"downloads"}'

2. 标签推荐接口
curl -X POST http://127.0.0.1:5000/recommend_tags -H "Content-Type: application/json" -d '{"user_id":"1","tags":["动作","科幻"]}'

3. 用户兴趣标签管理（添加标签）
curl -X POST http://127.0.0.1:5000/user_tags -H "Content-Type: application/json" -d '{"user_id":"1","tags":["动作","科幻"]}'

4. 用户兴趣标签管理（查询标签）
curl "http://127.0.0.1:5000/user_tags?user_id=1"

5. 用户兴趣标签管理（删除标签）
curl -X DELETE http://127.0.0.1:5000/user_tags -H "Content-Type: application/json" -d '{"user_id":"1","tags":["动作","科幻"]}'

6. 协同过滤推荐
curl -X POST http://127.0.0.1:5000/user_based_recommend -H "Content-Type: application/json" -d '{"user_id":"user1","top_n":3}'

7. Word2Vec状态检查
curl "http://127.0.0.1:5000/word2vec_status"

8. 调试接口（临时）
curl -X POST http://127.0.0.1:5000/debug_search -H "Content-Type: application/json" -d '{"keyword":"电影"}'

所有接口均可用Postman按上述参数测试。
"""

if __name__ == "__main__":
    try:
        logger.info("搜索推荐服务启动中...")
        app.run(host="0.0.0.0", port=5000)
    except Exception as e:
        logger.error(f"启动异常: {e}")
        import traceback
        traceback.print_exc()