A hybrid recommender approach based on Widrow-Hoff learning

Recommender is a personalized service in the adaptive information system, and it can provide personalized information according to individual information needs. As one of the known technology in the field of the recommender systems, collaborative filtering has been widely used in E-Commerce for its advantages. But the rating prediction mechanism of pure collaborative filtering is merely based on the ratings for visited items, and this limits its precision improvement. In this paper, we propose a refined hybrid recommender approach based on Widrow-Hoff learning algorithm. The proposed approach employs Widrow-Hoff algorithm to learn each user's profile from the contents of rated items, to improve the granularity of the user profiling. With the refined user profiles, collaborative filtering is employed to compute more precise similarity of different users, and predicts the ratings for unrated items. The improvement of performance is demonstrated by the experimental evaluation.