Abstract:

Collaborative filtering is one of the most successful techniques in Ecommerce recommender system. However, it faces severe problems of sparse user ratings and low recommendation accuracy. To solve the problems of lower recommendation quality caused by rating data sparseness and single rating, we propose a collaborative filtering recommendation algorithm based on ratings of item attributes. Firstly, we construct useritem attribute rating matrices using the mean value method or scaling method to transform single rating to multirating. Based on each rating matrix of attributes, we then calculate the similarity among users to obtain the preference set of the nearestneighbors, and accomplish a primary prediction for each set of the nearestneighbors based on useritem rating matrices. Finally, we calculate the weighted sum of multiple primary predictions as the final scores, and then complete the recommendation. The experimental results on the extended datasets of Movie Lens show that the proposed algorithm can get higher recommendation accuracy than traditional algorithms.

Key words: attribute rating;mean value method;scaling method;collaborative filtering;recommendation