Empirical study of recommender systems using linear classifiers

Abstract

Recommender systems use historical data on user prefer- ences and other available data on users (e.g., demographics) and items (e.g., taxonomy) to predict items a new user might like. Applications of these methods include recommending items for purchase and per- sonalizing the browsing experience on a web-site. Collaborative filter- ing methods have focused on using just the history of user preferences to make the recommendations. These methods have been categorized as memory-based if they operate over the entire data to make predic- tions and as model-based if they use the data to build a model which is then used for predictions. In this paper, we propose the use of lin- ear classifiers in a model-based recommender system. We compare our method with another model-based method using decision trees and with memory-based methods using data from various domains. Our experi- mental results indicate that these linear models are well suited for this application. They outperform the commonly proposed approach using a memory-based method in accuracy and also have a better tradeoff be- tween off-line and on-line computational requirements.