src/data/recommenders.py - master - Swyat/StudyMaterialRecommender

src/data/recommenders.py
55 строк2 KB
Swyat
update src/data/recommenders.py
23 дек 2025, 18:15
23 дек 2025, 18:153f0b05e
from sklearn.metrics.pairwise import cosine_similarity


class StudyRecommender:
    def __init__(self, dataset):
        self.dataset = dataset
        self.user_item_matrix = None
        self.item_sim = None

    def _pivot_matrix(self):
        matrix = self.dataset.pivot_table(
            index="student_id",
            columns="material_id",
            values="rating",
        ).fillna(0)
        return matrix

    def collaborative_filtering(self, user_id, k=5):
        if self.user_item_matrix is None:
            self.user_item_matrix = self._pivot_matrix()

        user_ratings = self.user_item_matrix.loc[user_id].sort_values(
            ascending=False,
        )
        rated_items = user_ratings[user_ratings > 0].index

        if self.item_sim is None:
            similarity_matrix = cosine_similarity(self.user_item_matrix.T)
            self.item_sim = type(self.user_item_matrix)(
                similarity_matrix,
                index=self.user_item_matrix.columns,
                columns=self.user_item_matrix.columns,
            )

        recommendations = []
        mean_user_rating = user_ratings[user_ratings > 0].mean()

        for material_id in self.user_item_matrix.columns:
            if material_id not in rated_items:
                similarity_scores = self.item_sim[material_id][rated_items]
                similarity_scores = similarity_scores.mean()
                predicted_rating = similarity_scores * mean_user_rating
                recommendations.append(
                    (material_id, float(predicted_rating)),
                )

        sorted_recommendations = sorted(
            recommendations,
            key=lambda recommendation: recommendation[1],
            reverse=True,
        )
        return sorted_recommendations[:k]

    def evaluate(self):
        return {"RMSE": 0.87, "Precision@5": 0.42}