recommend/recommend.py

import os
import time
import jieba
import fasttext
import pandas as pd
from flask import Flask, request, jsonify
from sqlalchemy import create_engine
from scipy.sparse import coo_matrix
from sklearn.metrics.pairwise import cosine_similarity
import pickle

app = Flask(__name__)

# === ✅ SQLAlchemy 数据库连接 ===
engine = create_engine("mysql+pymysql://sy:sy_password@49.233.215.144:3306/pt_station")

# === ✅ 加载 fastText 模型 ===
fasttext_model_path = 'E:\\course\\pt\\recommend\\models\\cc.zh.300.bin'
if not os.path.exists(fasttext_model_path):
    raise FileNotFoundError("fastText 模型文件不存在，请检查路径。")
print("加载 fastText 模型中...")
ft_model = fasttext.load_model(fasttext_model_path)
print("模型加载完成 ✅")

# === ✅ 用户标签行为矩阵构建 ===
def get_user_tag_matrix():
    df = pd.read_sql("SELECT user_id, tag, score FROM user_tag_scores", engine)
    print(df)
    df['user_id'] = df['user_id'].astype(str)
    user_map = {u: i for i, u in enumerate(df['user_id'].unique())}
    tag_map = {t: i for i, t in enumerate(df['tag'].unique())}
    df['user_index'] = df['user_id'].map(user_map)
    df['tag_index'] = df['tag'].map(tag_map)
    matrix = df.pivot_table(index='user_id', columns='tag', values='score', fill_value=0)
    sparse_matrix = coo_matrix((df['score'], (df['tag_index'], df['user_index'])))
    return df, matrix, sparse_matrix, user_map, tag_map

# === ✅ 基于 fastText 的语义相似推荐方法 ===
def semantic_recommend(user_id, topn=5):
    print(f"正在为用户 {user_id} 生成推荐...")

    # 读取数据库中的用户标签数据
    df = pd.read_sql("SELECT user_id, tag, score FROM user_tag_scores", engine)
    print(f"总记录数: {len(df)}")
    print(f"数据示例:\n{df.head()}")
    print(df.dtypes)
    user_id = str(user_id)  # 确保匹配

    # 获取该用户的所有标签（按分数从高到低排序）
    user_tags = df[df['user_id'] == user_id].sort_values(by="score", ascending=False)['tag'].tolist()
    print(f"用户 {user_id} 的标签（按分数排序）: {user_tags}")

    if not user_tags:
        print(f"用户 {user_id} 没有标签记录，返回空推荐结果。")
        return []

    # 截取前 3 个标签作为“兴趣标签”
    user_tags = user_tags[:3]
    print(f"用户 {user_id} 的 Top 3 标签: {user_tags}")

    # 构造所有标签的词向量
    all_tags = df['tag'].unique()
    print(f"所有唯一标签数量: {len(all_tags)}")

    tag_vectors = {}
    for tag in all_tags:
        vec = ft_model.get_word_vector(tag)
        tag_vectors[tag] = vec

    # 计算未出现过标签的相似度得分
    scores = {}
    for tag in all_tags:
        if tag in user_tags:
            continue
        vec = tag_vectors[tag]
        sim_total = 0.0
        for t in user_tags:
            sim = cosine_similarity([vec], [ft_model.get_word_vector(t)])[0][0]
            print(f"标签 [{tag}] 与用户标签 [{t}] 的相似度: {sim:.4f}")
            sim_total += sim
        avg_score = sim_total / len(user_tags)
        scores[tag] = avg_score
        print(f"标签 [{tag}] 的平均相似度得分: {avg_score:.4f}")

    # 排序并返回 topN 标签
    sorted_tags = sorted(scores.items(), key=lambda x: x[1], reverse=True)[:topn]
    print(f"\n最终推荐标签（前 {topn}）:")
    for tag, score in sorted_tags:
        print(f"{tag}: {score:.4f}")

    return [tag for tag, _ in sorted_tags]

# === ✅ ItemCF 推荐方法 ===
import os
import pickle

def itemcf_recommend(user_id, matrix, sim_path="./models/itemcf_sim.pkl", topn=5):
    user_id = str(user_id)  # 确保 user_id 类型一致
    print(matrix.index.dtype)
    print(type(user_id))  # 应该是 str

    if user_id not in matrix.index:
        print(f"⚠️ 用户 {user_id} 不在评分矩阵中。")
        return []

    if not os.path.exists(sim_path):
        print(f"⚠️ 用户 {user_id} 不在评分矩阵中。")
        train_and_save_itemcf()

    with open(sim_path, "rb") as f:
        sim_df = pickle.load(f)

    user_row = matrix.loc[user_id]
    user_tags = user_row[user_row > 0]

    if user_tags.empty:
        print(f"⚠️ 用户 {user_id} 没有任何标签评分记录。")
        return []

    print(f"用户 {user_id} 的标签评分:\n{user_tags}")

    scores = {}
    for tag, val in user_tags.items():
        if tag not in sim_df:
            print(f"标签 {tag} 在相似度矩阵中不存在，跳过。")
            continue
        sims = sim_df[tag].drop(index=user_tags.index, errors="ignore")
        for sim_tag, sim_score in sims.items():
            scores[sim_tag] = scores.get(sim_tag, 0) + sim_score * val

    if not scores:
        print(f"⚠️ 用户 {user_id} 无法生成推荐，可能是标签相似度不足。")
        return []

    sorted_tags = sorted(scores.items(), key=lambda x: x[1], reverse=True)
    print(f"推荐得分（前{topn}）:\n", sorted_tags[:topn])

    return [tag for tag, _ in sorted_tags[:topn]]


# === ✅ ItemCF 相似度训练 ===
def train_and_save_itemcf(path="./models/itemcf_sim.pkl"):
    _, matrix, _, _, _ = get_user_tag_matrix()
    tag_sim = cosine_similarity(matrix.T)
    sim_df = pd.DataFrame(tag_sim, index=matrix.columns, columns=matrix.columns)
    with open(path, "wb") as f:
        pickle.dump(sim_df, f)
    print("ItemCF 相似度矩阵已保存 ✅")

# === ✅ Flask 推荐接口 ===
import random

@app.route("/recommend_torrents", methods=["POST"])
def recommend_torrents():
    data = request.get_json()
    user_id = data.get("user_id")

    if not user_id:
        return jsonify({"error": "缺少 user_id"}), 400

    df, matrix, _, _, _ = get_user_tag_matrix()

    # 获取推荐标签
    itemcf_result = itemcf_recommend(user_id, matrix)
    semantic_result = semantic_recommend(user_id)


    print(f"ItemCF 推荐标签: {itemcf_result}")
    print(f"Semantic 推荐标签: {semantic_result}")

    all_tags = df['tag'].unique().tolist()

    # 存储标签及其推荐得分
    combined = []
    used_tags = set()

    def add_unique_tags(tags, method_name):
        for tag in tags:
            if tag not in used_tags:
                random_priority = random.uniform(0, 1)
                if method_name == 'ItemCF':
                    combined.append((tag, 'ItemCF', random_priority))
                elif method_name == 'Semantic':
                    combined.append((tag, 'Semantic', random_priority))
                used_tags.add(tag)

    # 添加 ItemCF 和 Semantic 推荐
    add_unique_tags(itemcf_result, 'ItemCF')
    add_unique_tags(semantic_result, 'Semantic')

    # 添加随机标签
    random.shuffle(all_tags)
    add_unique_tags(all_tags, 'Random')

    # 排序：按推荐得分排序，加入的随机值也会影响排序
    combined.sort(key=lambda x: x[2], reverse=True)

    # 根据标签获取种子 ID
    final_tags = [tag for tag, _, _ in combined]
    print(f"最终推荐标签: {final_tags}")
    torrent_ids = get_torrent_ids_by_tags(final_tags)

    return jsonify({"torrent_ids": torrent_ids})



from sqlalchemy.sql import text

import random
from sqlalchemy import text

def get_torrent_ids_by_tags(tags, limit_per_tag=10):
    if not tags:
        tags = []

    recommended_ids = set()
    with engine.connect() as conn:
        for tag in tags:
            query = text("""
                SELECT torrent_id
                FROM bt_torrent_tags 
                WHERE tag = :tag 
                LIMIT :limit
            """)
            result = conn.execute(query, {"tag": tag, "limit": limit_per_tag})
            for row in result:
                recommended_ids.add(row[0])

        # 获取数据库中所有 torrent_id
        all_query = text("SELECT DISTINCT torrent_id FROM bt_torrent_tags")
        all_result = conn.execute(all_query)
        all_ids = set(row[0] for row in all_result)

    # 剩下的（非推荐）种子 ID
    remaining_ids = all_ids - recommended_ids

    # 随机打乱推荐和剩下的 ID
    recommended_list = list(recommended_ids)
    remaining_list = list(remaining_ids)
    random.shuffle(recommended_list)
    random.shuffle(remaining_list)

    return recommended_list + remaining_list


# === ✅ 启动服务 ===
if __name__ == '__main__':
    train_and_save_itemcf()
    from waitress import serve
    serve(app, host="0.0.0.0", port=5000, threads=16)