Integrating Tensor Factorization with Neighborhood for Item Recommendation in Multidimensional Context

Item recommendation for multidimensional data context is getting increasing attention in recent years. Tensor factorization and neighborhood based collaborative filtering are the major techniques in use, but they address the item recommendation task for multidimensional data in quite different ways and have different strengths. In this paper, we discuss the characteristics of the two techniques, and present an approach for user profiling and neighborhood formation using multidimensional data, and also propose a novel collaborative filtering recommendation model which integrates the tensor factorization based and the neighborhood based collaborative filtering techniques for item recommendation with the Social Tagging Systems (STS) as the application domain. Meanwhile, the proposed recommendation approach is applicable to other application domains where multidimensional data is available. We empirically compare the proposed model against some state-of-the-art collaborative filtering recommendation approaches on two real-world datasets: Bibsonomy and MovieLens. The experimental results show the superiority of the proposed model in terms of recommendation quality.