diff --git a/recommend/inference.py b/recommend/inference.py
index 9601230..e2b7b00 100644
--- a/recommend/inference.py
+++ b/recommend/inference.py
@@ -1,32 +1,110 @@
 import sys
 sys.path.append('./')
 
-import time
 import torch
-import numpy as np
-from os import path
+import pymysql
+from flask_cors import CORS
+from flask import Flask, request, jsonify
 from model.LightGCN import LightGCN
 from utils.parse_args import args
 from utils.data_loader import EdgeListData
-from utils.data_generator import build_user_seed_graph
+from utils.graph_build import build_user_seed_graph
+
+app = Flask(__name__)
+CORS(app)
 
 args.device = 'cuda:7'
 args.data_path = './user_seed_graph.txt'
 args.pre_model_path = './model/LightGCN_pretrained.pt'
 
-def run_inference(user_id=1):
-    # 1. 实时生成user-seed交互图
-    print("正在生成用户-种子交互文件...")
-    build_user_seed_graph()
+# 数据库连接配置
+DB_CONFIG = {
+    'host': '10.126.59.25',
+    'port': 3306,
+    'user': 'root',
+    'password': '123456',
+    'database': 'pt_database_test',
+    'charset': 'utf8mb4'
+}
 
-    # 2. 加载数据集
-    print("正在加载数据集...")
-    t_data_start = time.time()
+TOPK = 2  # 默认推荐数量
+
+def user_cold_start(topk=TOPK):
+    """
+    冷启动：直接返回热度最高的topk个种子详细信息
+    """
+    conn = pymysql.connect(**DB_CONFIG)
+    cursor = conn.cursor()
+
+    # 查询热度最高的topk个种子
+    cursor.execute(
+        f"SELECT seed_id, owner_user_id, tags, title, size, popularity FROM Seed ORDER BY popularity DESC LIMIT %s",
+        (topk,)
+    )
+    seed_rows = cursor.fetchall()
+    seed_ids = [row[0] for row in seed_rows]
+    seed_map = {row[0]: row for row in seed_rows}
+
+    # 查询用户信息
+    owner_ids = list(set(row[1] for row in seed_rows))
+    if owner_ids:
+        format_strings_user = ','.join(['%s'] * len(owner_ids))
+        cursor.execute(
+            f"SELECT user_id, username FROM User WHERE user_id IN ({format_strings_user})",
+            tuple(owner_ids)
+        )
+        user_rows = cursor.fetchall()
+        user_map = {row[0]: row[1] for row in user_rows}
+    else:
+        user_map = {}
+
+    # 查询促销信息
+    if seed_ids:
+        format_strings = ','.join(['%s'] * len(seed_ids))
+        cursor.execute(
+            f"SELECT seed_id, discount FROM SeedPromotion WHERE seed_id IN ({format_strings})",
+            tuple(seed_ids)
+        )
+        promo_rows = cursor.fetchall()
+        promo_map = {row[0]: row[1] for row in promo_rows}
+    else:
+        promo_map = {}
+
+    cursor.close()
+    conn.close()
+
+    seed_list = []
+    for seed_id in seed_ids:
+        row = seed_map.get(seed_id)
+        if not row:
+            continue
+        owner_user_id = row[1]
+        seed_list.append({
+            'seed_id': seed_id,
+            'tags': row[2],
+            'title': row[3],
+            'size': row[4],
+            'username': user_map.get(owner_user_id, ""),
+            'popularity': row[5],
+            'discount': promo_map.get(seed_id, 1)
+        })
+    return seed_list
+
+def run_inference(user_id, topk=TOPK):
+    """
+    输入: user_id, topk
+    输出: 推荐的topk个种子ID列表（原始种子ID字符串）
+    """
+    user2idx, seed2idx = build_user_seed_graph(return_mapping=True)
+    idx2seed = {v: k for k, v in seed2idx.items()}
+
+    if user_id not in user2idx:
+        # 冷启动
+        return user_cold_start(topk)
+
+    user_idx = user2idx[user_id]
+
     dataset = EdgeListData(args.data_path, args.data_path)
-    t_data_end = time.time()
-
-    # 3. 加载LightGCN模型
-    print("正在加载模型参数...")
     pretrained_dict = torch.load(args.pre_model_path, map_location=args.device, weights_only=True)
     pretrained_dict['user_embedding'] = pretrained_dict['user_embedding'][:dataset.num_users]
     pretrained_dict['item_embedding'] = pretrained_dict['item_embedding'][:dataset.num_items]
@@ -35,23 +113,90 @@
     model.load_state_dict(pretrained_dict, strict=False)
     model.eval()
 
-    # 4. 推理
-    print(f"正在为用户 {user_id} 推理推荐结果...")
-    t_infer_start = time.time()
     with torch.no_grad():
         user_emb, item_emb = model.generate()
-        user_vec = user_emb[user_id].unsqueeze(0)
+        user_vec = user_emb[user_idx].unsqueeze(0)
         scores = model.rating(user_vec, item_emb).squeeze(0)
-        pred_item = torch.argmax(scores).item()
-    t_infer_end = time.time()
+        topk_indices = torch.topk(scores, topk).indices.cpu().numpy()
+        topk_seed_ids = [idx2seed[idx] for idx in topk_indices]
 
-    print(f"用户{user_id}下一个最可能点击的物品ID为: {pred_item}")
-    print(f"加载数据集耗时: {t_data_end - t_data_start:.4f} 秒")
-    print(f"推理耗时: {t_infer_end - t_infer_start:.4f} 秒")
+    return topk_seed_ids
+
+def seed_info(topk_seed_ids):
+    """
+    输入: topk_seed_ids（种子ID字符串列表）
+    输出: 推荐种子的详细信息列表，每个元素为dict
+    """
+    if not topk_seed_ids:
+        return []
+
+    conn = pymysql.connect(**DB_CONFIG)
+    cursor = conn.cursor()
+
+    # 查询种子基本信息
+    format_strings = ','.join(['%s'] * len(topk_seed_ids))
+    cursor.execute(
+        f"SELECT seed_id, owner_user_id, tags, title, size, popularity FROM Seed WHERE seed_id IN ({format_strings})",
+        tuple(topk_seed_ids)
+    )
+    seed_rows = cursor.fetchall()
+    seed_map = {row[0]: row for row in seed_rows}
+
+    # 查询用户信息
+    owner_ids = list(set(row[1] for row in seed_rows))
+    if owner_ids:
+        format_strings_user = ','.join(['%s'] * len(owner_ids))
+        cursor.execute(
+            f"SELECT user_id, username FROM User WHERE user_id IN ({format_strings_user})",
+            tuple(owner_ids)
+        )
+        user_rows = cursor.fetchall()
+        user_map = {row[0]: row[1] for row in user_rows}
+    else:
+        user_map = {}
+
+    # 查询促销信息
+    cursor.execute(
+        f"SELECT seed_id, discount FROM SeedPromotion WHERE seed_id IN ({format_strings})",
+        tuple(topk_seed_ids)
+    )
+    promo_rows = cursor.fetchall()
+    promo_map = {row[0]: row[1] for row in promo_rows}
+
+    cursor.close()
+    conn.close()
+
+    seed_list = []
+    for seed_id in topk_seed_ids:
+        row = seed_map.get(seed_id)
+        if not row:
+            continue
+        owner_user_id = row[1]
+        seed_list.append({
+            'seed_id': seed_id,
+            'tags': row[2],
+            'title': row[3],
+            'size': row[4],
+            'username': user_map.get(owner_user_id, ""),
+            'popularity': row[5],
+            'discount': promo_map.get(seed_id, 1)
+        })
+    return seed_list
+
+@app.route('/recommend', methods=['POST'])
+def recommend():
+    data = request.get_json()
+    user_id = data.get('user_id')
+    try:
+        result = run_inference(user_id)
+        # 如果是冷启动直接返回详细信息，否则查详情
+        if isinstance(result, list) and result and isinstance(result[0], dict):
+            seed_list = result
+        else:
+            seed_list = seed_info(result)
+        return jsonify({'recommend': seed_list})
+    except Exception as e:
+        return jsonify({'error': str(e)}), 400
 
 if __name__ == "__main__":
-    t_start = time.time()
-    user_id = 1
-    run_inference(user_id)
-    t_end = time.time()
-    print(f"脚本总耗时: {t_end - t_start:.4f} 秒")
\ No newline at end of file
+    app.run(host='0.0.0.0', port=5000)