推荐系统：基于用户的协同过滤

所用数据：https://grouplens.org/datasets/movielens/

数据读取

"""
@author: LiShiHang
@software: PyCharm
@file: data_preprocessing.py
@time: 2019/4/3 21:12
@desc:
"""
import pandas as pd
import numpy as np

np.set_printoptions(threshold=np.inf)


def read_data(dir="../data/ml-1m/"):
    """
    将data/ml-1m文件夹中的dat文件转化为csv
    :param dir:
    :return:
    """
    ####################################
    f_users = pd.read_table(
        dir + "users.dat",
        engine="python",
        sep="::",
        names=[
            'userID',
            'Gender',
            'Age',
            'Occupation',
            'Zip-code'])
    f_users.to_csv("users.csv", index=False)
    f_users.info()
    ###########################################

    f_ratings = pd.read_table(dir + "ratings.dat", engine="python", sep="::",
                              names=['UserID', 'MovieID', 'Rating', 'Timestamp'])
    f_ratings.to_csv("ratings.csv", index=False)
    f_ratings.info()

    ################################################
    f_movies = pd.read_table(dir + "movies.dat", engine="python", sep="::",
                             names=['MovieID', 'Title', 'Genres'])
    f_movies.to_csv("movies.csv", index=False)
    f_movies.info()

    print("finish.")

if __name__ == '__main__':

    read_data()

模型构建

"""
@author: LiShiHang
@software: PyCharm
@file: user_cf.py
@time: 2019/4/4 8:31
@desc:
"""
import math
import data_preprocessing
import pandas as pd


class UserCF():

    def __init__(self, path_rating):

        self.data_rating = pd.read_csv(path_rating)

    def calc_user_sim(self, item1, item2):
        """
        计算相似度
        :param item1:
        :param item2:
        :return:
        """
        cosine = len(set(item1) & set(item2)) / \
            math.sqrt(len(item1) * len(item2))
        return cosine

    def get_users_topN(self, source_user_id, N):
        """
        得到N个相似用户
        :param source_user_id:
        :param topN:
        :return: [[用户ID，相似度]]
        """
        source_user_movies = self.data_rating[self.data_rating['UserID']
                                              == source_user_id]["MovieID"]  # 目标用户看过的电影ID

        others_id = [
            i for i in set(
                self.data_rating['UserID']) if i != source_user_id]  # 其他用户ID

        others_movies = [
            self.data_rating[self.data_rating['UserID'] == i]["MovieID"] for i in others_id]  # [[其他用户看的电影ID] 其他用户ID]

        sim_list = [
            self.calc_user_sim(
                source_user_movies,
                movies) for movies in others_movies]  # 根据目标用户和其他用户 看过的电影ID计算相似度

        sim_list = sorted(zip(others_id, sim_list),
                          key=lambda x: x[1], reverse=True)

        return sim_list[:N]

    def get_candidate(self, source_user_id):
        """
        得到候选电影清单
        :param source_user_id:
        :return:
        """
        source_user_movies = set(
            self.data_rating[self.data_rating['UserID'] == source_user_id]["MovieID"])  # 目标用户看过的电影ID

        others_movies = set(
            self.data_rating[self.data_rating['UserID'] != source_user_id]["MovieID"])

        candidate_movies = others_movies - source_user_movies

        return candidate_movies

    def get_item_topN(self, top_n_users, candidate_movies, topN):
        """
        得到推荐电影列表
        :param top_n_users:
        :param candidate_movies:
        :param topN:
        :return:
        """
        top_n_users_data = [self.data_rating[self.data_rating['UserID'] == i]
                            for i, _ in top_n_users]  # 相似用户数据

        interest_item = []

        for cm in candidate_movies:  # 候选电影
            tmp = []
            for user_data in top_n_users_data:  # 相似用户，[[用户ID，相似度]]
                if cm in user_data["MovieID"].values:
                    tmp.append(user_data[user_data["MovieID"]
                                         == cm]['Rating'].values[0] / 5.0)
                else:
                    tmp.append(0)
            interest = sum([top_n_users[i][1] * tmp[i]
                            for i in range(len(top_n_users))])  # 相似用户对每个候选电影的感兴趣度（评分）
            interest_item.append((cm, interest))

        interest_item = sorted(interest_item, key=lambda x: x[1], reverse=True)

        return interest_item[:topN]


if __name__ == '__main__':

    ucf = UserCF("ratings.csv")

    ui = 1
    top_n_users = ucf.get_users_topN(ui, 10)

    candidate = ucf.get_candidate(ui)

    top_n_movies = ucf.get_item_topN(top_n_users, candidate, 10)

    print(top_n_movies)  # 推荐的电影ID，推荐程度

    # 显示电影名
    movies = pd.read_csv("movies.csv")
    # print("*" * 20)
    # for i in ucf.data_rating[ucf.data_rating["UserID"]==ui]["MovieID"]: # 目标用户看过的电影名称
    #     print(*movies[movies["MovieID"]==i].values[0])
    print("*" * 20)
    for i, j in top_n_movies:  # 推荐看的电影名称
        print(*movies[movies["MovieID"] == i].values[0], j)

由于最近较忙，博客不做说明。

参考链接：
https://blog.csdn.net/sinat_33741547/article/category/6442592
https://github.com/lpty/recommendation
https://www.cnblogs.com/tbiiann/
https://github.com/apachecn/RecommenderSystems/tree/master/py3.x