Abstract:

Addressing the information overloading problem, the collaborative filtering is an effective technique, and extensively applied in recommender systems. It make predictions by finding users with similar taste or items that have been similarly chosen. However, as the number of users or items grows rapidly, the traditional collaborative filtering approach is suffering from the data sparsity problem. The sparse user-item associations can generate inaccurate neighborhood for each user or item. A distributed hybrid collaborative filtering method was proposed based on Map Reduce, aiming at improving the recommendation quality. This method utilizes user features and ratings to construct item preference vectors. Then, it clusters items using fuzzy K-Means algorithm, and respectively chooses similar items from each clustering, finally it combines all predictions from each clustering and makes recommendation. Experiments show that the distributed hybrid collaborative filtering method can help reduce the sparsity problem, and improve the recommendation accuracy.

Key words: distributed framework, personalized recommendation, collaborative filtering, fuzzy clustering