基于卷积神经网络CNN的电影推荐系统

本项目使用文本卷积神经网络，并使用MovieLens数据集完成电影推荐的任务。

推荐系统在日常的网络应用中无处不在，比如网上购物、网上买书、新闻app、社交网络、音乐网站、电影网站等等等等，有人的地方就有推荐。根据个人的喜好，相同喜好人群的习惯等信息进行个性化的内容推荐。比如打开新闻类的app，因为有了个性化的内容，每个人看到的新闻首页都是不一样的。

这当然是很有用的，在信息爆炸的今天，获取信息的途径和方式多种多样，人们花费时间最多的不再是去哪获取信息，而是要在众多的信息中寻找自己感兴趣的，这就是信息超载问题。为了解决这个问题，推荐系统应运而生。

协同过滤是推荐系统应用较广泛的技术，该方法搜集用户的历史记录、个人喜好等信息，计算与其他用户的相似度，利用相似用户的评价来预测目标用户对特定项目的喜好程度。优点是会给用户推荐未浏览过的项目，缺点呢，对于新用户来说，没有任何与商品的交互记录和个人喜好等信息，存在冷启动问题，导致模型无法找到相似的用户或商品。

为了解决冷启动的问题，通常的做法是对于刚注册的用户，要求用户先选择自己感兴趣的话题、群组、商品、性格、喜欢的音乐类型等信息，比如豆瓣FM：

下载数据集

运行下面代码把数据集下载下来

import pandas as pd
from sklearn.model_selection import train_test_split
import numpy as np
from collections import Counter
import tensorflow as tf
 
import os
import pickle
import re
from tensorflow.python.ops import math_ops

from urllib.request import urlretrieve
from os.path import isfile, isdir
from tqdm import tqdm
import zipfile
import hashlib
 
def _unzip(save_path, _, database_name, data_path):
    """
    解压
    :param save_path: The path of the gzip files
    :param database_name: Name of database
    :param data_path: Path to extract to
    :param _: HACK - Used to have to same interface as _ungzip
    """
    print('Extracting {}...'.format(database_name))
    with zipfile.ZipFile(save_path) as zf:
        zf.extractall(data_path)
 
def download_extract(database_name, data_path):
    """
    下载提取数据
    :param database_name: Database name
    """
    DATASET_ML1M = 'ml-1m'
 
    if database_name == DATASET_ML1M:
        url = 'http://files.grouplens.org/datasets/movielens/ml-1m.zip'
        hash_code = 'c4d9eecfca2ab87c1945afe126590906'
        extract_path = os.path.join(data_path, 'ml-1m')
        save_path = os.path.join(data_path, 'ml-1m.zip')
        extract_fn = _unzip
 
    if os.path.exists(extract_path):
        print('Found {} Data'.format(database_name))
        return
 
    if not os.path.exists(data_path):
        os.makedirs(data_path)
 
    if not os.path.exists(save_path):
        with DLProgress(unit='B', unit_scale=True, miniters=1, desc='Downloading {}'.format(database_name)) as pbar:
            urlretrieve(
                url,
                save_path,
                pbar.hook)
 
    assert hashlib.md5(open(save_path, 'rb').read()).hexdigest() == hash_code, \
        '{} file is corrupted.  Remove the file and try again.'.format(save_path)
 
    os.makedirs(extract_path)
    try:
        extract_fn(save_path, extract_path, database_name, data_path)
    except Exception as err:
        shutil.rmtree(extract_path)  # Remove extraction folder if there is an error
        raise err
 
    print('Done.')
    # Remove compressed data
#     os.remove(save_path)
 
class DLProgress(tqdm):
    """
    下载时处理进度条
    """
    last_block = 0
 
    def hook(self, block_num=1, block_size=1, total_size=None):
        """
        A hook function that will be called once on establishment of the network connection and
        once after each block read thereafter.
        :param block_num: A count of blocks transferred so far
        :param block_size: Block size in bytes
        :param total_size: The total size of the file. This may be -1 on older FTP servers which do not return
                            a file size in response to a retrieval request.
        """
        self.total = total_size
        self.update((block_num - self.last_block) * block_size)
        self.last_block = block_num

data_dir = './'
download_extract('ml-1m', data_dir)

Extracting ml-1m...
Done.

先来看看数据

本项目使用的是MovieLens 1M 数据集，包含6000个用户在近4000部电影上的1亿条评论。

数据集分为三个文件：

• 用户数据users.dat
• 电影数据movies.dat
• 评分数据ratings.dat

用户数据

• 用户ID
• 性别
• 年龄
• 职业ID
• 邮编

数据中的格式：UserID::Gender::Age::Occupation::Zip-code

• Gender is denoted by a "M" for male and "F" for female

• Age is chosen from the following ranges:

  • 1: "Under 18"
  • 18: "18-24"
  • 25: "25-34"
  • 35: "35-44"
  • 45: "45-49"
  • 50: "50-55"
  • 56: "56+"

• Occupation is chosen from the following choices:

  • 0: "other" or not specified
  • 1: "academic/educator"
  • 2: "artist"
  • 3: "clerical/admin"
  • 4: "college/grad student"
  • 5: "customer service"
  • 6: "doctor/health care"
  • 7: "executive/managerial"
  • 8: "farmer"
  • 9: "homemaker"
  • 10: "K-12 student"
  • 11: "lawyer"
  • 12: "programmer"
  • 13: "retired"
  • 14: "sales/marketing"
  • 15: "scientist"
  • 16: "self-employed"
  • 17: "technician/engineer"
  • 18: "tradesman/craftsman"
  • 19: "unemployed"
  • 20: "writer"

users_title = ['UserID', 'Gender', 'Age', 'OccupationID', 'Zip-code']
users = pd.read_table('./ml-1m/users.dat', sep='::', header=None, names=users_title, engine = 'python')
users.head()

	UserID	Gender	Age	OccupationID	Zip-code
0	1	F	1	10	48067
1	2	M	56	16	70072
2	3	M	25	15	55117
3	4	M	45	7	02460
4	5	M	25	20	55455

可以看出UserID、Gender、Age和Occupation都是类别字段，其中邮编字段是我们不使用的。

电影数据

• 电影ID
• 电影名
• 电影风格

数据中的格式：MovieID::Title::Genres

• Titles are identical to titles provided by the IMDB (including
  year of release)

• Genres are pipe-separated and are selected from the following genres:

  • Action
  • Adventure
  • Animation
  • Children's
  • Comedy
  • Crime
  • Documentary
  • Drama
  • Fantasy
  • Film-Noir
  • Horror
  • Musical
  • Mystery
  • Romance
  • Sci-Fi
  • Thriller
  • War
  • Western

movies_title = ['MovieID', 'Title', 'Genres']
movies = pd.read_table('./ml-1m/movies.dat', sep='::', header=None, names=movies_title, engine = 'python')
movies.head()

	MovieID	Title	Genres
0	1	Toy Story (1995)	Animation|Children's|Comedy
1	2	Jumanji (1995)	Adventure|Children's|Fantasy
2	3	Grumpier Old Men (1995)	Comedy|Romance
3	4	Waiting to Exhale (1995)	Comedy|Drama
4	5	Father of the Bride Part II (1995)	Comedy

MovieID是类别字段，Title是文本，Genres也是类别字段

评分数据

• 用户ID
• 电影ID
• 评分
• 时间戳

数据中的格式：UserID::MovieID::Rating::Timestamp

• UserIDs range between 1 and 6040
• MovieIDs range between 1 and 3952
• Ratings are made on a 5-star scale (whole-star ratings only)
• Timestamp is represented in seconds since the epoch as returned by time(2)
• Each user has at least 20 ratings

ratings_title = ['UserID','MovieID', 'Rating', 'timestamps']
ratings = pd.read_table('./ml-1m/ratings.dat', sep='::', header=None, names=ratings_title, engine = 'python')
ratings.head()

	UserID	MovieID	Rating	timestamps
0	1	1193	5	978300760
1	1	661	3	978302109
2	1	914	3	978301968
3	1	3408	4	978300275
4	1	2355	5	978824291

评分字段Rating就是我们要学习的targets，时间戳字段我们不使用。

来说说数据预处理

• UserID、Occupation和MovieID不用变。
• Gender字段：需要将‘F’和‘M’转换成0和1。
• Age字段：要转成7个连续数字0~6。
• Genres字段：是分类字段，要转成数字。首先将Genres中的类别转成字符串到数字的字典，然后再将每个电影的Genres字段转成数字列表，因为有些电影是多个Genres的组合。
• Title字段：处理方式跟Genres字段一样，首先创建文本到数字的字典，然后将Title中的描述转成数字的列表。另外Title中的年份也需要去掉。
• Genres和Title字段需要将长度统一，这样在神经网络中方便处理。空白部分用‘< PAD >’对应的数字填充。

实现数据预处理

def load_data():
    """
    从文件中加载数据集
    """
    # 读取User数据
    users_title = ['UserID', 'Gender', 'Age', 'JobID', 'Zip-code']
    users = pd.read_table('./ml-1m/users.dat', sep='::', header=None, names=users_title, engine = 'python')
    users = users.filter(regex='UserID|Gender|Age|JobID')
    users_orig = users.values
    
    # 改变User数据中性别和年龄
    gender_map = {'F':0, 'M':1}
    users['Gender'] = users['Gender'].map(gender_map)
 
    age_map = {val:ii for ii,val in enumerate(set(users['Age']))}
    users['Age'] = users['Age'].map(age_map)
 
    # 读取Movie数据集
    movies_title = ['MovieID', 'Title', 'Genres']
    movies = pd.read_table('./ml-1m/movies.dat', sep='::', header=None, names=movies_title, engine = 'python')
    movies_orig = movies.values
    # 将Title中的年份去掉
    pattern = re.compile(r'^(.*)\((\d+)\)$')
 
    title_map = {val:pattern.match(val).group(1) for ii,val in enumerate(set(movies['Title']))}
    movies['Title'] = movies['Title'].map(title_map)
 
    # 电影类型转数字字典
    genres_set = set()
    for val in movies['Genres'].str.split('|'):
        genres_set.update(val)
 
    genres_set.add('<PAD>')
    genres2int = {val:ii for ii, val in enumerate(genres_set)}
 
    # 将电影类型转成等长数字列表，长度是18
    genres_map = {val:[genres2int[row] for row in val.split('|')] for ii,val in enumerate(set(movies['Genres']))}
 
    for key in genres_map:
        for cnt in range(max(genres2int.values()) - len(genres_map[key])):
            genres_map[key].insert(len(genres_map[key]) + cnt,genres2int['<PAD>'])
    
    movies['Genres'] = movies['Genres'].map(genres_map)
 
    # 电影Title转数字字典
    title_set = set()
    for val in movies['Title'].str.split():
        title_set.update(val)
    
    title_set.add('<PAD>')
    title2int = {val:ii for ii, val in enumerate(title_set)}
 
    # 将电影Title转成等长数字列表，长度是15
    title_count = 15
    title_map = {val:[title2int[row] for row in val.split()] for ii,val in enumerate(set(movies['Title']))}
    
    for key in title_map:
        for cnt in range(title_count - len(title_map[key])):
            title_map[key].insert(len(title_map[key]) + cnt,title2int['<PAD>'])
    
    movies['Title'] = movies['Title'].map(title_map)
 
    # 读取评分数据集
    ratings_title = ['UserID','MovieID', 'ratings', 'timestamps']
    ratings = pd.read_table('./ml-1m/ratings.dat', sep='::', header=None, names=ratings_title, engine = 'python')
    ratings = ratings.filter(regex='UserID|MovieID|ratings')
 
    # 合并三个表
    data = pd.merge(pd.merge(ratings, users), movies)
    
    # 将数据分成X和y两张表
    target_fields = ['ratings']
    features_pd, targets_pd = data.drop(target_fields, axis=1), data[target_fields]
    
    features = features_pd.values
    targets_values = targets_pd.values
    
    return title_count, title_set, genres2int, features, targets_values, ratings, users, movies, data, movies_orig, users_orig

加载数据并保存到本地

• title_count：Title字段的长度（15）
• title_set：Title文本的集合
• genres2int：电影类型转数字的字典
• features：是输入X
• targets_values：是学习目标y
• ratings：评分数据集的Pandas对象
• users：用户数据集的Pandas对象
• movies：电影数据的Pandas对象
• data：三个数据集组合在一起的Pandas对象
• movies_orig：没有做数据处理的原始电影数据
• users_orig：没有做数据处理的原始用户数据

# 加载数据
title_count, title_set, genres2int, features, targets_values, ratings, users, movies, data, movies_orig, users_orig = load_data()
 
# 存入文件中
pickle.dump((title_count, title_set, genres2int, features, targets_values, ratings, users, movies, data, movies_orig, users_orig), open('preprocess.p', 'wb'))

预处理后的数据

users.head()

	UserID	Gender	Age	JobID
0	1	0	0	10
1	2	1	5	16
2	3	1	6	15
3	4	1	2	7
4	5	1	6	20

movies.head()

	MovieID	Title	Genres
0	1	[310, 2184, 634, 634, 634, 634, 634, 634, 634,...	[0, 18, 7, 17, 17, 17, 17, 17, 17, 17, 17, 17,...
1	2	[1182, 634, 634, 634, 634, 634, 634, 634, 634,...	[3, 18, 8, 17, 17, 17, 17, 17, 17, 17, 17, 17,...
2	3	[5011, 4744, 2629, 634, 634, 634, 634, 634, 63...	[7, 9, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17,...
3	4	[4095, 1535, 1886, 634, 634, 634, 634, 634, 63...	[7, 5, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17,...
4	5	[3563, 1725, 3790, 3727, 838, 343, 634, 634, 6...	[7, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17...

movies.values[0]

array([1,
       list([310, 2184, 634, 634, 634, 634, 634, 634, 634, 634, 634, 634, 634, 634, 634]),
       list([0, 18, 7, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17])],
      dtype=object)

从本地读取数据

title_count, title_set, genres2int, features, targets_values, ratings, users, movies, data, movies_orig, users_orig = pickle.load(open('preprocess.p', mode='rb'))

模型设计

通过研究数据集中的字段类型，我们发现有一些是类别字段，通常的处理是将这些字段转成one hot编码，但是像UserID、MovieID这样的字段就会变成非常的稀疏，输入的维度急剧膨胀，这是我们不愿意见到的，毕竟我这小笔记本不像大厂动辄能处理数以亿计维度的输入：）

所以在预处理数据时将这些字段转成了数字，我们用这个数字当做嵌入矩阵的索引，在网络的第一层使用了嵌入层，维度是（N，32）和（N，16）。

电影类型的处理要多一步，有时一个电影有多个电影类型，这样从嵌入矩阵索引出来是一个（n，32）的矩阵，因为有多个类型嘛，我们要将这个矩阵求和，变成（1，32）的向量。

电影名的处理比较特殊，没有使用循环神经网络，而是用了文本卷积网络，下文会进行说明。

从嵌入层索引出特征以后，将各特征传入全连接层，将输出再次传入全连接层，最终分别得到（1，200）的用户特征和电影特征两个特征向量。

我们的目的就是要训练出用户特征和电影特征，在实现推荐功能时使用。得到这两个特征以后，就可以选择任意的方式来拟合评分了。我使用了两种方式，一个是上图中画出的将两个特征做向量乘法，将结果与真实评分做回归，采用MSE优化损失。因为本质上这是一个回归问题，另一种方式是，将两个特征作为输入，再次传入全连接层，输出一个值，将输出值回归到真实评分，采用MSE优化损失。

实际上第二个方式的MSE loss在0.8附近，第一个方式在1附近，5次迭代的结果。

文本卷积网络

网络看起来像下面这样

图片来自Kim Yoon的论文：Convolutional Neural Networks for Sentence Classification

将卷积神经网络用于文本的文章建议你阅读Understanding Convolutional Neural Networks for NLP

网络的第一层是词嵌入层，由每一个单词的嵌入向量组成的嵌入矩阵。下一层使用多个不同尺寸（窗口大小）的卷积核在嵌入矩阵上做卷积，窗口大小指的是每次卷积覆盖几个单词。这里跟对图像做卷积不太一样，图像的卷积通常用2x2、3x3、5x5之类的尺寸，而文本卷积要覆盖整个单词的嵌入向量，所以尺寸是（单词数，向量维度），比如每次滑动3个，4个或者5个单词。第三层网络是max pooling得到一个长向量，最后使用dropout做正则化，最终得到了电影Title的特征。

辅助函数

import tensorflow as tf
import os
import pickle
 
def save_params(params):
    """
    保存参数到文件中
    """
    pickle.dump(params, open('params.p', 'wb'))
 
 
def load_params():
    """
    从文件中加载参数
    """
    return pickle.load(open('params.p', mode='rb'))

编码实现

# 嵌入矩阵的维度
embed_dim = 32
# 用户ID个数
uid_max = max(features.take(0,1)) + 1 # 6040
# 性别个数
gender_max = max(features.take(2,1)) + 1 # 1 + 1 = 2
# 年龄类别个数
age_max = max(features.take(3,1)) + 1 # 6 + 1 = 7
# 职业个数
job_max = max(features.take(4,1)) + 1# 20 + 1 = 21
 
# 电影ID个数
movie_id_max = max(features.take(1,1)) + 1 # 3952
# 电影类型个数
movie_categories_max = max(genres2int.values()) + 1 # 18 + 1 = 19
# 电影名单词个数
movie_title_max = len(title_set) # 5216
 
# 对电影类型嵌入向量做加和操作的标志，考虑过使用mean做平均，但是没实现mean
combiner = "sum"
 
# 电影名长度
sentences_size = title_count # = 15
# 文本卷积滑动窗口，分别滑动2, 3, 4, 5个单词
window_sizes = {2, 3, 4, 5}
# 文本卷积核数量
filter_num = 8
 
# 电影ID转下标的字典，数据集中电影ID跟下标不一致，比如第5行的数据电影ID不一定是5
movieid2idx = {val[0]:i for i, val in enumerate(movies.values)}

超参

# Number of Epochs
num_epochs = 5
# Batch Size
batch_size = 256
 
dropout_keep = 0.5
# Learning Rate
learning_rate = 0.0001
# Show stats for every n number of batches
show_every_n_batches = 20
 
save_dir = './save'

输入

定义输入的占位符

def get_inputs():
    uid = tf.placeholder(tf.int32, [None, 1], name="uid")
    user_gender = tf.placeholder(tf.int32, [None, 1], name="user_gender")
    user_age = tf.placeholder(tf.int32, [None, 1], name="user_age")
    user_job = tf.placeholder(tf.int32, [None, 1], name="user_job")
    
    movie_id = tf.placeholder(tf.int32, [None, 1], name="movie_id")
    movie_categories = tf.placeholder(tf.int32, [None, 18], name="movie_categories")
    movie_titles = tf.placeholder(tf.int32, [None, 15], name="movie_titles")
    targets = tf.placeholder(tf.int32, [None, 1], name="targets")
    LearningRate = tf.placeholder(tf.float32, name = "LearningRate")
    dropout_keep_prob = tf.placeholder(tf.float32, name = "dropout_keep_prob")
    return uid, user_gender, user_age, user_job, movie_id, movie_categories, movie_titles, targets, LearningRate, dropout_keep_prob

构建神经网络

定义User的嵌入矩阵

def get_user_embedding(uid, user_gender, user_age, user_job):
    with tf.name_scope("user_embedding"):
        uid_embed_matrix = tf.Variable(tf.random_uniform([uid_max, embed_dim], -1, 1), name = "uid_embed_matrix")
        uid_embed_layer = tf.nn.embedding_lookup(uid_embed_matrix, uid, name = "uid_embed_layer")
    
        gender_embed_matrix = tf.Variable(tf.random_uniform([gender_max, embed_dim // 2], -1, 1), name= "gender_embed_matrix")
        gender_embed_layer = tf.nn.embedding_lookup(gender_embed_matrix, user_gender, name = "gender_embed_layer")
        
        age_embed_matrix = tf.Variable(tf.random_uniform([age_max, embed_dim // 2], -1, 1), name="age_embed_matrix")
        age_embed_layer = tf.nn.embedding_lookup(age_embed_matrix, user_age, name="age_embed_layer")
        
        job_embed_matrix = tf.Variable(tf.random_uniform([job_max, embed_dim // 2], -1, 1), name = "job_embed_matrix")
        job_embed_layer = tf.nn.embedding_lookup(job_embed_matrix, user_job, name = "job_embed_layer")
    return uid_embed_layer, gender_embed_layer, age_embed_layer, job_embed_layer

将User的嵌入矩阵一起全连接生成User的特征

def get_user_feature_layer(uid_embed_layer, gender_embed_layer, age_embed_layer, job_embed_layer):
    with tf.name_scope("user_fc"):
        #第一层全连接
        uid_fc_layer = tf.layers.dense(uid_embed_layer, embed_dim, name = "uid_fc_layer", activation=tf.nn.relu)
        gender_fc_layer = tf.layers.dense(gender_embed_layer, embed_dim, name = "gender_fc_layer", activation=tf.nn.relu)
        age_fc_layer = tf.layers.dense(age_embed_layer, embed_dim, name ="age_fc_layer", activation=tf.nn.relu)
        job_fc_layer = tf.layers.dense(job_embed_layer, embed_dim, name = "job_fc_layer", activation=tf.nn.relu)
        
        #第二层全连接
        user_combine_layer = tf.concat([uid_fc_layer, gender_fc_layer, age_fc_layer, job_fc_layer], 2)  #(?, 1, 128)
        user_combine_layer = tf.contrib.layers.fully_connected(user_combine_layer, 200, tf.tanh)  #(?, 1, 200)
    
        user_combine_layer_flat = tf.reshape(user_combine_layer, [-1, 200])
    return user_combine_layer, user_combine_layer_flat

定义Movie ID的嵌入矩阵

def get_movie_id_embed_layer(movie_id):
    with tf.name_scope("movie_embedding"):
        movie_id_embed_matrix = tf.Variable(tf.random_uniform([movie_id_max, embed_dim], -1, 1), name = "movie_id_embed_matrix")
        movie_id_embed_layer = tf.nn.embedding_lookup(movie_id_embed_matrix, movie_id, name = "movie_id_embed_layer")
    return movie_id_embed_layer

对电影类型的多个嵌入向量做加和

def get_movie_categories_layers(movie_categories):
    with tf.name_scope("movie_categories_layers"):
        movie_categories_embed_matrix = tf.Variable(tf.random_uniform([movie_categories_max, embed_dim], -1, 1), name = "movie_categories_embed_matrix")
        movie_categories_embed_layer = tf.nn.embedding_lookup(movie_categories_embed_matrix, movie_categories, name = "movie_categories_embed_layer")
        if combiner == "sum":
            movie_categories_embed_layer = tf.reduce_sum(movie_categories_embed_layer, axis=1, keep_dims=True)
    #     elif combiner == "mean":
 
    return movie_categories_embed_layer

Movie Title的文本卷积网络实现

def get_movie_cnn_layer(movie_titles):
    #从嵌入矩阵中得到电影名对应的各个单词的嵌入向量
    with tf.name_scope("movie_embedding"):
        movie_title_embed_matrix = tf.Variable(tf.random_uniform([movie_title_ma