CollaborativeFiltering.py

from surprise import SVD, Dataset, Reader
from surprise.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder

class CollaborativeFiltering:
    def __init__(self, user_behavior, games):
        self.user_behavior = user_behavior
        self.games = games
        self.model = SVD()
        self.user_encoder = LabelEncoder()
        self.game_encoder = LabelEncoder()
        self.df_long = None
        self.trainset = None
        self._prepare_data()
        self._train_model()

    def _prepare_data(self):
        self.user_behavior = self.user_behavior.pivot_table(index=["user-id", "game_title"], columns="action", values="value", fill_value=0).reset_index()
        merged_data = self.user_behavior.merge(self.games, left_on="game_title", right_on="name", how="left")
        merged_data.dropna(subset=["appid", "name"], inplace=True)
        merged_data["user_id_encoded"] = self.user_encoder.fit_transform(merged_data["user-id"])
        merged_data["game_id_encoded"] = self.game_encoder.fit_transform(merged_data["game_title"])
        self.df_long = merged_data[["user_id_encoded", "game_id_encoded", "play"]]

    def _train_model(self):
        reader = Reader(rating_scale=(self.df_long["play"].min(), self.df_long["play"].max()))
        data = Dataset.load_from_df(self.df_long, reader)
        self.trainset, _ = train_test_split(data, test_size=0.2)
        self.model.fit(self.trainset)

    def recommend_games(self, user_id, top_n=5):
        user_games = self.df_long[self.df_long["user_id_encoded"] == user_id]["game_id_encoded"].unique()
        all_games = self.df_long["game_id_encoded"].unique()
        games_to_predict = [game for game in all_games if game not in user_games]
        predictions = [self.model.predict(user_id, game) for game in games_to_predict]
        predictions.sort(key=lambda x: x.est, reverse=True)
        top_games = [pred.iid for pred in predictions[:top_n]]
        return self.game_encoder.inverse_transform(top_games)