Abstract: In collaborative filtering, the recommendation performance is limited by the data sparsity of rating matrix. To alleviate this problem, a rough set based collaborative filtering method is proposed, which is helpful to reduce the influence of the data sparsity to some extent. First, the procedure of data pre-processing is executed. As the eigenvalue for user classification, users ratings with the rating number are employed. Then, an attribute reduction approach of rough set is introduced to eliminate the item which has little effect with respect to user classification, and a smaller scale user-item rating matrix is generated to lower the data sparsity. Finally, the similarity between users is calculated based on the reducing attributes to obtain real neighbors of active users. The experimental results demonstrate that the proposed algorithm in MAE, COVERAGE, PRECISION and RECALL are significantly better than the ones of COS, PCC, ADCOS, NHSM.

Key words: collaborative filtering, attribute reduction, data preprocess, classification label, rough set