基于协同过滤和内容过滤的steam游戏推荐模型

import pandas as pd import numpy as np import pyhdfs import translators as ts import os import joblib from sklearn.neighbors import NearestNeighbors from sklearn.feature_extraction.text import TfidfVectorizer from sklearn.metrics.pairwise import cosine_similarity from scipy.sparse import coo_matrix from utils.uploads import * # 构建游戏推荐模型 class GameRecommendation: def __init__(self,recommendations_data,games_data): self.games_data=games_data[['app_id','title']].copy() self.knn_model=NearestNeighbors(metric='euclidean', algorithm='brute') self.user_ids=recommendations_data['user_id'].astype('category').cat.codes.astype('int32') self.item_ids=recommendations_data['app_id'].astype('category').cat.codes.astype('int32') self.unique_user_ids=recommendations_data['user_id'].astype('category').cat.categories.astype('int32') self.unique_item_ids=recommendations_data['app_id'].astype('category').cat.categories.astype('int32') self.user_game_matrix=coo_matrix((recommendations_data['score'], (self.user_ids, self.item_ids))).astype('float32') self.tf_tags=TfidfVectorizer(analyzer='word',ngram_range=(1, 1),min_df=1, stop_words='english') self.tf_description=TfidfVectorizer(analyzer='word',ngram_range=(1, 2),min_df=1, stop_words='english') self.tags_matrix=self.tf_tags.fit_transform(games_data['tags']).astype('float32') self.description_matrix=self.tf_description.fit_transform(games_data['description']).astype('float32') self.cosine_tags=None self.cosine_description=None def train(self): self.knn_model.fit(self.user_game_matrix) def train_tags(self): self.cosine_tags=cosine_similarity(self.tags_matrix,self.tags_matrix) def train_description(self): self.cosine_description=cosine_similarity(self.description_matrix,self.description_matrix) # 根据游戏名称获取app_id再获取该游戏在user_game_matrix中的索引 def get_app_index(self,game): app_id = self.games_data[self.games_data['title'] == game]['app_id'].values[0] app_index = self.unique_item_ids.get_indexer([app_id])[0] return app_index # 获取相似用户 def get_similar_users(self,games,scores,n_neighbors=6): app_index=[self.get_app_index(game) for game in games] user = np.zeros(self.user_game_matrix.shape[1]) for i in range(len(app_index)): user[app_index[i]] = scores[i] user = user.reshape(1, -1) _, indices = self.knn_model.kneighbors(user, n_neighbors=n_neighbors) similar_users = [self.unique_user_ids[i] for i in indices.flatten()[1:]] return similar_users # 获取相似用户的游戏 def get_users_games(self,similar_users): users_games = set() for user in similar_users: user_index = self.unique_user_ids.get_indexer([user])[0] user_games = self.user_game_matrix.getrow(user_index).nonzero()[1] for game_id in user_games: users_games.add(self.games_data[self.games_data['app_id'] == self.unique_item_ids[game_id]]['title'].values[0]) return users_games # 获取相似游戏 def get_similar_games(self,games): games_index=[] for game in games: games_index.append(self.games_data[self.games_data['title'] == game].index[0]) recommended_games=pd.DataFrame() for game_index in games_index: similar_tags = self.cosine_tags[game_index].argsort()[:-10:-1] similar_descrip = self.cosine_description[game_index].argsort()[:-2:-1] similar_games = [self.games_data['title'].iloc[i] for i in similar_tags] similar_games.extend([self.games_data['title'].iloc[i] for i in similar_descrip]) recommended_games['title']=similar_games recommended_games['score']=np.arange(len(similar_tags),-1,-1) # 去除重复的游戏 recommended_games.drop_duplicates('title',keep='first',inplace=True) # 根据score排序，保留前5个 recommended_games=recommended_games.sort_values('score',ascending=False).head(5) return set(recommended_games['title'].values) # 推荐游戏 def recommend(self,games,scores): similar_users=self.get_similar_users(games,scores) users_games=self.get_users_games(similar_users) recommend_games=self.get_similar_games(users_games) return recommend_games # 推荐游戏的中文翻译 def recommend_CN(self,games,scores): similar_users=self.get_similar_users(games,scores) users_games=self.get_users_games(similar_users) recommend_games=self.get_similar_games(users_games) recommend_games_CN=set() for game in recommend_games: recommend_games_CN.add(ts.translate_text(game, to_language='zh-CN')) return recommend_games_CN fs=pyhdfs.HdfsClient(hosts='192.168.88.132:9870',user_name='zjy') # 加载数据 recommendations_data=get_recommendations_data(fs,'/input/recommendations/') games_data=pd.read_csv(fs.open('/uploads/games_metadata.csv')) recommendations_data.head() games_data.head() model=GameRecommendation(recommendations_data,games_data) model.train() model.train_tags() model.train_description() games=['Prince of Persia: Warrior Within™'] scores=[100] recommend_games=model.recommend(games,scores) recommend_games games=['Dungeons & Dragons: Dark Alliance','Dungeon of the ENDLESS™','Labyrinth of Refrain: Coven of Dusk'] scores=[100,20,50] recommend_games=model.recommend(games,scores) recommend_games recommend_games_CN=model.recommend_CN(games,scores) recommend_games_CN # 保存模型,如果文件夹不存在则创建 if not os.path.exists('output/model/'): os.makedirs('output/model/') joblib.dump(model, 'output/model/model.pkl')