Multi-clustering Used as Neighbourhood Identification Strategy in Recommender Systems

Abstract

This article describes clustering approach to neighbourhood calculation in collaborative filtering recommender systems. Precise identification of neighbours of an active object (a user to whom recommendations are generated) is very important due to its direct impact on quality of generated recommendation lists. Clustering techniques, although improving time effectiveness of recommender systems, can negatively affect quality (precision) of recommendations. In this article it is proposed a new algorithm based on multi-clustering, as well as author’s description of this term. Despite of various definitions in papers, a common distinctive feature of multi-clustering is its multiple point of view of one dataset. Various views discover their different aspects, selecting the most appropriate data model to solution of a current problem. The article contains experiments confirming advantage of multi-clustering approach over the traditional method based on single-scheme clustering. The results include recommendation quality and time effectiveness comparison, as well.