An effective preprocessing algorithm for model building in collaborative filtering-based recommender system. (12th April 2019)

• Record Type: Journal Article
• Authors: Srikanth, T.; Shashi, M.
• Is Part Of: International journal of business intelligence and data mining. Volume 14:Number 4(2019)
• Page Start: 489
• Page End: 503
• Publication Date: 2019-04-12
• Subjects: recommender system -- collaborative filtering -- dimensionality reduction -- pre-processing -- sparsity -- scalability -- low rank approximation; 006.312
• Languages: English
• ISSNs: 1743-8187

Abstract

Recommender systems suggest interesting items for online users based on the ratings expressed by them for the other items maintained globally as the rating matrix. The rating matrix is often sparse and very huge due to large number of users expressing their ratings only for a few items among the large number of alternatives. Sparsity and scalability are the challenging issues to achieve accurate predictions in recommender systems. This paper focuses on model building approach to collaborative filtering-based recommender systems using low rank matrix approximation algorithms for achieving scalability and accuracy while dealing with sparse rating matrices. A novel preprocessing methodology is proposed to counter data sparsity problem by transforming the sparse rating matrix denser before extracting latent factors to appropriately characterise the users and items in low dimensional space. The quality of predictions made either directly or indirectly through user clustering were investigated and found to be competitive with the existing collaborative filtering methods in terms of reduced MAE and increased NDCG values on bench mark datasets.