电影推荐系统：TMDB 5000 Movie Dataset（一）

image.png

一.数据集简介
tmdb_5000_movies和tmdb_5000_credits这两个数据集所包含的特征
tmdb_5000_movies：
homepage，id，original_title，overview，popularity，production_companies，production_countries，release_date，spoken_languages，status，tagline，vote_average

tmdb_5000_credits：
movie_id，title，cast，crew

二.对原始数据集进行预处理

import json
import pandas as pd
#load_tmdb_movies用于载入movie数据集
def load_tmdb_movies(path):
    df = pd.read_csv(path)
    df['release_date'] = pd.to_datetime(df['release_date']).apply(lambda x: x.date())
    json_columns = ['genres', 'keywords', 'production_countries',
                    'production_companies', 'spoken_languages']
    for column in json_columns:
        df[column] = df[column].apply(json.loads)
    return df

#load_tmdb_credits用于载入credits数据集
def load_tmdb_credits(path):
    df = pd.read_csv(path)
    json_columns = ['cast', 'crew']
    for column in json_columns:
        df[column] = df[column].apply(json.loads)
    return df

#用于更改原始数据中的列名
TMDB_TO_IMDB_SIMPLE_EQUIVALENCIES = {
    'budget': 'budget',
    'genres': 'genres',
    'revenue': 'gross',
    'title': 'movie_title',
    'runtime': 'duration',
    'original_language': 'language',
    'keywords': 'plot_keywords',
    'vote_count': 'num_voted_users'}

#检索函数
def safe_access(container, index_values):
    result = container
    try:
        for idx in index_values:
            result = result[idx]
        return result
    except IndexError or KeyError:
        return pd.np.nan

#关键字处理函数，可以将关键字用"|"隔开
def pipe_flatten_names(keywords):
    return '|'.join([x['name'] for x in keywords])

#对原始数据
def convert_to_original_format(movies, credits):
    tmdb_movies = movies.copy()
    tmdb_movies.rename(columns=TMDB_TO_IMDB_SIMPLE_EQUIVALENCIES, inplace=True)
    tmdb_movies['title_year'] = pd.to_datetime(tmdb_movies['release_date']).apply(lambda x: x.year)
    tmdb_movies['country'] = tmdb_movies['production_countries'].apply(lambda x: safe_access(x, [0, 'name']))
    tmdb_movies['language'] = tmdb_movies['spoken_languages'].apply(lambda x: safe_access(x, [0, 'name']))
    tmdb_movies['director_name'] = credits['crew'].apply(get_director)
    tmdb_movies['actor_1_name'] = credits['cast'].apply(lambda x: safe_access(x, [1, 'name']))
    tmdb_movies['actor_2_name'] = credits['cast'].apply(lambda x: safe_access(x, [2, 'name']))
    tmdb_movies['actor_3_name'] = credits['cast'].apply(lambda x: safe_access(x, [3, 'name']))
    tmdb_movies['genres'] = tmdb_movies['genres'].apply(pipe_flatten_names)
    tmdb_movies['plot_keywords'] = tmdb_movies['plot_keywords'].apply(pipe_flatten_names)
    return tmdb_movies

#导入必要的包
import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
import seaborn as sns
import math, nltk, warnings
from nltk.corpus import wordnet
from sklearn import linear_model
from sklearn.neighbors import NearestNeighbors
from fuzzywuzzy import fuzz
from wordcloud import WordCloud, STOPWORDS
plt.rcParams["patch.force_edgecolor"] = True
plt.style.use('fivethirtyeight')
mpl.rc('patch', edgecolor = 'dimgray', linewidth=1)
from IPython.core.interactiveshell import InteractiveShell
InteractiveShell.ast_node_interactivity = "last_expr"
pd.options.display.max_columns = 50
%matplotlib inline
warnings.filterwarnings('ignore')
PS = nltk.stem.PorterStemmer()

#载入原始数据并进行预处理
credits = load_tmdb_credits("tmdb_5000_credits.csv")
movies = load_tmdb_movies("tmdb_5000_movies.csv")
df_initial = convert_to_original_format(movies, credits)
print('Shape:',df_initial.shape)

tab_info = pd.DataFrame(df_initial.dtypes).T.rename(index={0: 'column type'})
tab_info = tab_info.append(pd.DataFrame(df_initial.isnull().sum()).T.rename(index={0:'null values'}))
tab_info=tab_info.append(pd.DataFrame(df_initial.isnull().sum()/df_initial.shape[0]*100).T. rename(index={0:'null values (%)'}))

从输出结果中可以看出df_initial的shape为(4803, 26)。另外，tab_info表示为（因为是截图所以还有好几列被省略了）：

image.png

下一篇将会进入正式的推荐系统设计环节~