An Improved Collaborative Filtering Algorithm Combining User Activity Level

Abstract

Collaborative filtering (CF), which plays an important role in making personalized recommendation, is one of the most traditional and effective recommendation algorithms. However, there are several factors that impact its recommendation accuracy, e.g., the sparse matrix problem. In the past studies, most researchers merely focused on user ratings to model user profile but ignored the implying patterns. In this paper, we utilize user activity to discriminate user rating patterns and propose a new method of user-based collaborative filtering based on user activity level. Experimental results on movie-lens data-set has proved that the algorithm we proposed improves recommendation accuracy significantly compared with traditional user-based CF algorithm with respect to various evaluation metrics.