A Bicriteria Clustering Approach for Collaborative Filtering

Clustering is one of the essential methods of data reduction. It is possible to find homogenous sub-sets of huge amount of data by employing clustering. In collaborative filtering schemes, clustering is used to form off-line user or item neighborhoods in order to enhance online performance. Classical clustering methods for collaborative filtering are only based on distances or correlations among entities. Thus, it is hard to form neighborhoods without sacrificing any useful entity by clustering. In this paper, we introduce a new bicriteria k-means clustering approach for collaborative filtering. We employ a degree of uncertainty of users along with similarities in order to obtain a single clustering criterion. We perform experiments on two benchmark data sets in order to measure the proposed approach's accuracy. Experimental outcomes indicate that, it is possible to improve accuracy of a recommender system using bicriteria-based k-means clustering.