An efficient collaborative filtering approach using smoothing and fusing

Daqiang Zhang, Jiannong Cao, Jingyu Zhou, Minyi Guo, Vaskar Raychoudhury

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review

15 Citations (Scopus)

Abstract

Collaborative Filtering (CF) has achieved widespread success in recommender systems such as Amazon and Yahoo! music. However, CF usually suffers from two fundamental problems- data sparsity and limited scalability. Among the two broad classes of CF approaches, namely, memory-based and model-based, the former usually falls short of the system scalability demands, because these approaches predict user preferences over the entire item-user matrix. The latter often achieves unsatisfactory accuracy, because they cannot capture precisely the diversity in user rating styles. In this paper, we propose an efficient Collaborative Filtering approach using Smoothing and Fusing (CFSF) strategies. CFSF formulates the CF problem as a local prediction problem by mapping it from the entire large-scale item-user matrix to a locally reduced item-user matrix. Given an active item and a user, CFSF dynamically constructs a local item-user matrix as the basis of prediction. To alleviate data sparsity, CFSF presents a fusion strategy for the local item-user matrix that fuses ratings of the same user makes on similar items, and ratings of likeminded users make on the same and similar items. To eliminate diversity in user rating styles, CFSF uses a smoothing strategy that clusters users over the entire item-user matrix and then smoothes ratings within each user cluster. Empirical study shows that CFSF outperforms the state-of-the-art CF approaches in terms of both accuracy and scalability.
Original languageEnglish
Title of host publicationICPP-2009 - The 38th International Conference on Parallel Processing
Pages558-565
Number of pages8
DOIs
Publication statusPublished - 1 Dec 2009
Event38th International Conference on Parallel Processing, ICPP-2009 - Vienna, Austria
Duration: 22 Sep 200925 Sep 2009

Conference

Conference38th International Conference on Parallel Processing, ICPP-2009
Country/TerritoryAustria
CityVienna
Period22/09/0925/09/09

ASJC Scopus subject areas

  • Software
  • Mathematics(all)
  • Hardware and Architecture

Cite this