Indexed by:期刊论文

Date of Publication:2013-05-01

Journal:Journal of Computational Information Systems

Included Journals:EI、Scopus

Volume:9

Issue:9

Page Number:3759-3766

ISSN No.:15539105

Abstract:Collaborative filtering is one of the most successful techniques in recommender systems, but it encounters the challenge of rating sparsity. Although consumer interest diffusion mechanism can make some recommendations with few ratings, the diffusion-based collaborative filtering algorithms on consumer-product bipartite graph have to suffer the high computation complexity and sometimes cannot hit the unpopular products in which consumers might be interested. In this paper, a Consumer-ProductGroup diffusion model is developed. This model consists of both global and local analyzers. The global analyzer explores consumer's interests in different product groups which can be obtained offline through Self-Organizing Map clustering if there is no available information about product groups. The local analyzer predicts consumer's interests in products corresponding to the partitioned groups. Based on the developed model, a recommendation algorithm is proposed. As shown in the experimental results, the recommendation accuracy and diversity have been improved, meanwhile the computational cost keeps the lower level. ? 2013 by Binary Information Press.