Content-based Dimensionality Reduction for Recommender Systems

This paper constructs a feature profile of a user based on both collaborative and content features and applies Latent Semantic Indexing (LSI) to reveal the dominant features of a users, and provides recommendations according to this dimensionally-reduced feature profile.

Abstract

Recommender Systems are gaining widespread acceptance in e-commerce applications to confront the information overload problem. Collaborative Filtering (CF) is a successful recommendation technique, which is based on past ratings of users with similar preferences. In contrast, Content-based Filtering (CB) exploits information solely derived from document or item features (e.g. terms or attributes). CF has been combined with CB to improve the accuracy of recommendations. A major drawback in most of these hybrid approaches was that these two techniques were executed independently. In this paper, we construct a feature profile of a user based on both collaborative and content features. We apply Latent Semantic Indexing (LSI) to reveal the dominant features of a user. We provide recommendations according to this dimensionally-reduced feature profile. We perform experimental comparison of the proposed method against well-known CF, CB and hybrid algorithms. Our results show significant improvements in terms of providing accurate recommendations.