An empirical analysis on SVD based recommendation techniques
Rachana Mehta, Keyur Rana · 2017 Innovations in Power and Advanced Computing Technologies (i-PACT) · 2017
Recommender systems are used to accurately provide users with information or services of their interest. Collaborative filtering is a widely used recommendation approach. However, the data sparsity and cold-start user problems, often limit the recommendation performance and quality. To address such issues, matrix factorization techniques based on Singular Value Decomposition have well emerged. These techniques map users and items into low dimensional latent feature spaces, and use them for recommendation tasks. Matrix factorization techniques have evolved from utilizing simpler user-item rating information to utilizing auxiliary information of time or trust, to improvise the predictive accuracy of recommenders. In this paper, we present an empirical analysis on such techniques and explore the changes observed in recommendation performance by the usage of auxiliary information.