Reducing Data Sparsity in Recommender Systems

Abstract

Recommender systems are used to find user's interested things among a huge amount of digital information. Collaborative filtering is used to generate recommendations. However, the data sparsity problem leads to generate unreasonable recommendations for those users who provide no ratings. From this point, this paper presents a modest approach to enhance prediction in movielens dataset with high sparsity by applying collaborative filtering methods. The proposal consists of three consequence phases: preprocessing phase, similarity phase, prediction phase. The experimental results obtained conducting similarity measures against movielens user rating datasets show that the result of prediction is enhanced about 10% to15% with the non-sparse rating matrix.