推荐系统实践—ItemCF实现

参考：https://github.com/Lockvictor/MovieLens-RecSys/blob/master/usercf.py#L169

数据集

本文使用了MovieLens中的ml-100k小数据集，数据集的地址为：传送门
该数据集中包含了943个独立用户对1682部电影做的10000次评分。

完整代码

总体和UserCF差不多，将用户相似度的计算改为物品相似度的计算即可。

  1 import numpy as np
  2 import pandas as pd
  3 import math
  4 from collections import defaultdict
  5 from operator import itemgetter
  6 
  7 np.random.seed(1)
  8 
  9 
 10 class ItemCF(object):
 11 
 12     def __init__(self):
 13         self.train_set = {}
 14         self.test_set = {}
 15         self.movie_popularity = {}
 16 
 17         self.tot_movie = 0
 18         self.W = {}   # 相似度矩阵
 19 
 20         self.K = 160   # 最接近的K部电影
 21         self.M = 10   # 推荐电影数
 22 
 23     def split_data(self, data, ratio):
 24         ''' 按ratio的比例分成训练集和测试集 '''
 25         for line in data.itertuples():
 26             user, movie, rating = line[1], line[2], line[3]
 27             if np.random.random() < ratio:
 28                 self.train_set.setdefault(user, {})
 29                 self.train_set[user][movie] = int(rating)
 30             else:
 31                 self.test_set.setdefault(user, {})
 32                 self.test_set[user][movie] = int(rating)
 33         print('数据预处理完成')
 34 
 35     def item_similarity(self):
 36         ''' 计算物品相似度 '''
 37         for user, items in self.train_set.items():
 38             for movie in items.keys():
 39                 if movie not in self.movie_popularity:   # 用于后面计算新颖度
 40                     self.movie_popularity[movie] = 0
 41                 self.movie_popularity[movie] += 1
 42         self.tot_movie = len(self.movie_popularity)  # 用于计算覆盖率
 43 
 44         C, N = {}, {}    # C记录电影两两之间共同喜欢的人数， N记录电影的打分人数
 45         for user, items in self.train_set.items():
 46             for m1 in items.keys():
 47                 N.setdefault(m1, 0)
 48                 N[m1] += 1
 49                 C.setdefault(m1, defaultdict(int))
 50                 for m2 in items.keys():
 51                     if m1 == m2:
 52                         continue
 53                     else:
 54                         C[m1][m2] += 1
 55 
 56         count = 1
 57         for u, related_movies in C.items():
 58             print('\r相似度计算进度:{:.2f}%'.format(count * 100 / self.tot_movie), end='')
 59             count += 1
 60             self.W.setdefault(u, {})
 61             for v, cuv in related_movies.items():
 62                 self.W[u][v] = float(cuv) / math.sqrt(N[u] * N[v])
 63         print('\n相似度计算完成')
 64 
 65     def recommend(self, u):
 66         ''' 推荐M部电影 '''
 67         rank = {}
 68         user_movies = self.train_set[u]
 69 
 70         for movie, rating in user_movies.items():
 71             for related_movie, similarity in sorted(self.W[movie].items(), key=itemgetter(1), reverse=True)[0:self.K]:
 72                 if related_movie in user_movies:
 73                     continue
 74                 else:
 75                     rank.setdefault(related_movie, 0)
 76                     rank[related_movie] += similarity * rating
 77         return sorted(rank.items(), key=itemgetter(1), reverse=True)[0:self.M]
 78 
 79     def evaluate(self):
 80         ''' 评测算法 '''
 81         hit = 0
 82         ret = 0
 83         rec_tot = 0
 84         pre_tot = 0
 85         tot_rec_movies = set()  # 推荐电影
 86         for user in self.train_set:
 87             test_movies = self.test_set.get(user, {})
 88             rec_movies = self.recommend(user)
 89             for movie, pui in rec_movies:
 90                 if movie in test_movies.keys():
 91                     hit += 1
 92                 tot_rec_movies.add(movie)
 93                 ret += math.log(1+self.movie_popularity[movie])
 94             pre_tot += self.M
 95             rec_tot += len(test_movies)
 96         precision = hit / (1.0 * pre_tot)
 97         recall = hit / (1.0 * rec_tot)
 98         coverage = len(tot_rec_movies) / (1.0 * self.tot_movie)
 99         ret /= 1.0 * pre_tot
100         print('precision=%.4f' % precision)
101         print('recall=%.4f' % recall)
102         print('coverage=%.4f' % coverage)
103         print('popularity=%.4f' % ret)
104 
105 
106 if __name__ == '__main__':
107     data = pd.read_csv('u.data', sep='\t', names=['user_id', 'item_id', 'rating', 'timestamp'])
108     itemcf = ItemCF()
109     itemcf.split_data(data, 0.7)
110     itemcf.item_similarity()
111     itemcf.evaluate()

结果