A Multiple Feature Vector Based Interest Model And Its Application

Nowadays the Web is fooded with massive information. It becomes more andmore impossible for a person to manually extract interested information from the Web.To solve such problem, recommender systems are proposed and studied. A recom-mender system can recommend interested items to a user by analysing his/her prefer-ence. A recommender system usually needs to employ an interest model for learninguser preference or say user profle. Generally, an interest model have to solve the fol-lowing problems:1) how to represent a user’s interests.2) how to depict user interests’evolutionincludingeliminationofoldinterests, emergenceofnewinterests, distinctionbetween long-term and short-term interests and so on.3) how to update users’ inter-ests according to their feedback. In this thesis, we propose an interest model and tryto solve the above problems. The main contributions of this thesis are two-fold: frstlywe propose an interest model based on multiple feature vectors and secondly we im-prove the Rocchio algorithm for updating user profles. We conducted experiments onMovieLens dataset and the results prove our proposed interest model is efective andpractical.