Capturing user behavior in multimedia recommenders

Despite the great amount of work done in the last decade, the design of intelligent systems for effective browsing of large multimedia repositories still remains an open field. In this paper, we propose an extension of our previous multimedia recommender, and introduce mechanisms to improve the performance of the system for registered users who explicitly login before starting a browsing session, by dynamically capturing their browsing behavior and adjusting the recommendations accordingly. In addition, we introduce tracking of time spent by users on each item, and take this information into account when computing recommendations. Preliminary results are presented and compared with our previous system.