Evolution of Singular Value Decomposition in Recommendation Systems : A Review

Abstract

Proliferation of internet and web applications has led to exponential growth of users and information over web. In such information overload scenarios, recommender systems have shown their prominence by providing user with information of their interest. Recommender systems provide item recommendation or generate predictions. Amongst the various recommendation approaches, collaborative filtering techniques have emerged well because of its wide item applicability. Model-based collaborative filtering techniques which use parameterised model for prediction are more preferred as compared to their memory-based counterparts. However, the existing techniques deals with static data and are less accurate over sparse, high dimensional data. In order to alleviate such issues, matrix factorisation techniques like singular value decomposition are preferred. These techniques have evolved from using simple user-item rating information to auxiliary social and temporal information. In this paper, we provide a comprehensive review of such matrix factorisation techniques and their applicability to different input data.