关键词: LDA, 协同过滤, 聚类, 相似度计算, 时间因子

Abstract: Recommendation system is widely used in e-commerce, and collaborative filtering is one of the most successful techniques in the recommendation system. With the increasing of the e-commerce data, the problem of the sparsity of the user-item rating matrix becomes more and more obvious, which has become the bottleneck of the recommendation system. To improve the recommendation quality under the sparse dataset environment, this paper proposed an improved collaborative filtering algorithm based on LDA model. We first built LDA model according to the user-item rating matrix, and got user-item selection probability matrix. And then, we clustered the item set by item properties, and cut the matrix by cluster results. Finally, in the process of similarity calculation, we introduced time factor to improve similarity calculation formula. Experimental results on Movie Lens datasets show that the proposed model gets better performance than traditional collaborative filtering algorithm in MAE.

Key words: LDA, collaborative filtering, clustering, similarity calculation, time factor