EduCoder_Study_RS/shixun_recommend_online.py

import json
from datetime import datetime
from config import logger
from config import test_user_id
from ranking.shixun.user_recall_rank_features import init_rank_features
from ranking.shixun.user_recall_rank_features import build_rank_features_online
from matching.shixun.cold_start_recall import cold_start_user_recall
from matching.shixun.multi_recall_predict import multi_recall_predict

init_rank_features()

from ranking.shixun.xdeepfm_rank_predict import xdeepfm_ranker_predict
from ranking.shixun.bst_ranker_predict import bst_ranker_predict
from ranking.shixun.difm_ranker_predict import difm_ranker_predict

def shixun_recommend_online(user_id, disciplines_id_list=None, topk=10, rank_method='1'):
    """
    根据用户ID获取推荐的实训ID列表
    1. 使用多路召回获取用户的实训ID列表
    2. 根据召回列表生成排序模型特征
    3. 用排序模型对召回列表重新排序
    4. 返回排序后的结果
    """
    start_time = datetime.now()

    logger.info(f"本次需要进行推荐的用户ID: {user_id}")

    recommend_results = {}
    recommend_results.clear()    

    # 1.先进行多路召回
    recommend_results, only_cold_start_recall = multi_recall_predict(user_id, disciplines_id_list, topk=topk)

    if only_cold_start_recall and (len(recommend_results) > 0):
        # 计算耗时毫秒
        end_time = datetime.utcnow()
        cost_time_millisecond = round(float((end_time - start_time).microseconds / 1000.0), 3)    

        logger.info(f"本次推荐总耗时: {cost_time_millisecond} 毫秒")
        return recommend_results        

    # 2.构建排序模型特征
    user_item_feats_df = build_rank_features_online(user_id, recommend_results)

    # 3.使用排序模型对召回的候选物品进行排序
    # 如果没有召回数据则根据兴趣标签使用冷启动召回推荐
    if user_item_feats_df.empty:
        recommend_results = cold_start_user_recall(disciplines_id_list, topk=topk)
    else:
        if topk > user_item_feats_df.shape[0]:
            topk = user_item_feats_df.shape[0]

        if rank_method == '0':
            rank_results = xdeepfm_ranker_predict(user_item_feats_df, topk=topk)
        elif rank_method == '1':
            rank_results = bst_ranker_predict(user_item_feats_df, topk=topk)
        elif rank_method == '2':
            rank_results = difm_ranker_predict(user_item_feats_df, topk=topk)


        recommend_results = dict(zip(rank_results['shixun_id'], rank_results['shixun_name']))

    # 计算耗时毫秒
    end_time = datetime.utcnow()
    cost_time_millisecond = round(float((end_time - start_time).microseconds / 1000.0), 3)    

    logger.info(f"本次推荐总耗时: {cost_time_millisecond} 毫秒")
    return recommend_results


if __name__ == '__main__':
    recommend_results = shixun_recommend_online(user_id=test_user_id, 
                                                disciplines_id_list=[], 
                                                topk=10, 
                                                rank_method='1')
    print(json.dumps(recommend_results, ensure_ascii=False, indent=4))