Using Temporal Information in Collaborative Filtering: An Empirical Study

Collaborative filtering is a widely used and proven method of building recommender systems that provide personalized recommendations on products or services based on explicit ratings from users. Recommendation accuracy becomes an important factor in some e-commerce environments (such as a mobile environment as a result of limited connection time and device size). As user preferences change over time, temporal information can improve recommendation accuracy. In this paper, we present a variety of temporal information and investigate how such temporal information affects the accuracy of collaborative filtering-based recommender systems. The temporal information includes item launch time, user buying time, the time difference between the two, as well as several combinations of these three types of temporal information. We conducted several experiments on a collaborative filtering system for recommending character images (wallpapers) in a mobile e-commerce environment. Empirical results show that the effectiveness of temporal information depends on the type of items and the user group of a given e-commerce environment. Our findings give insights on how to incorporate temporal information to maximize the efficiency of collaborative filtering in various e-commerce environments.