Employing neighborhood reduction for alleviating sparsity and cold start problems in user-based collaborative filtering

Abstract

Recommender system (RS) can produce personalized service to users by analyzing their historical information. User-based collaborative filtering (UBCF) approach is widely utilized in practical RSs because of its excellent performance. However, the traditional UBCF suffers from several inherent problems, such as data sparsity and new user cold start. In this paper, we propose a novel approach, namely covering reduction collaborative filtering, to solve data sparsity and new user cold start problems in UBCF. First, we define the redundant users in a new user’s neighborhood through a detailed analysis on two real-world datasets (i.e., MovieLens and Netflix). Then, we analyze the intrinsic connection between redundant users in UBCF and redundant elements in covering-based rough sets, and transform the redundant user removal issue into the redundant element reduction. Furthermore, a cover is built for each new user according to the information of candidate neighbors. And the covering reduction algorithm is employed to remove the redundant elements in the cover of each new user, removing all reducible elements in a cover means redundant users in the neighborhood of a new user are removed. Finally, rating scores for unrated items are predicted by aggregating the ratings of remaining users after reduction. And items with the highest predicted rating scores will be recommended to the new user. Experimental results suggest that for the sparse datasets that often occur in real RSs, the proposed approach outperforms those of existing work and can provide recommendations for a new user with satisfactory accuracy and diversity simultaneously without requiring any other special additional information.