Finding users´ latent interests for recommendation by learning classifier systems

Collaborative filtering often used in e-commerce applications, is a method to cluster similar users based on their profiles, characteristics or attitudes on specific subjects. The paper proposes a novel method to implement dynamic collaborative filtering by genetics based machine learning, in which we employ a learning classifier system extended to multiple environments. The characteristics of the dynamic collaborative filtering method are summarized as follows: (1) it is effective in distributed computer environments with PCs even for a small number of users; (2) it learns users´ profiles from the individual behaviors and then generates recommendations and advice for each user; (3) the results are automatically accumulated in a local system on a PC, then they are distributed via smart IC cards while the users are interacting with the system. The method has been implemented and validated in the Group Trip Advisor prototype, a PC based distributed recommendation system for travel information