Clustering Approach for Multidimensional Recommender Systems

Abstract

Side information has been incorporated into traditional recommender systems to further enhance their performance, especially to alleviate the data sparsity and cold start issues. Side information in recommendations are the user-item related contents like user demographic data, movie genre, contextual or multi-criteria ratings. Incorporation of side information into classical recommender system often leads to multidimensionality problem, which imposes new challenges for the researchers. Therefore, the main objective of this work is to develop a side information based recommender system and handle multidimensionality issue to produce improved recommendations. The proposed approach is divided into three phases. In the first phase, user clusters are created using a side information clustering. In the second phase, top-K neighborhood set formed through intra-cluster distance computation using Mahalanobis distance measure. In the third phase, prediction and recommendations are generated for the users. Experimental results show the superiority of clustering based approach over non-clustering approach.