LEARNING USERS' DECISION-MAKING PATTERNS FOR IMPROVING RECOMMENDER SYSTEM

摘要

Recommendation, which selects the most relevant contents for users' preferences, is one of the decision support techniques. Content-based filtering (CBF) and collaborative filtering (CF) are widely used recommendation approaches. They only focus on whether the users would choose the recommendations or not. The current approaches do not concentrate on the problem of how the users select the favorite contents. This paper suggests an approach of learning users' decision making patterns in order to help the users to select favorite contents efficiently. The learning recommender methods have demerits such as cold start problem, which will be solved with our proposed fusion model in this study. The fusion model contains both learning methods and non-learning methods and reflects the users' decision-making patterns. For the effective recommendation, our proposed system have four user models: preference model, interest-contents group model, neighborhood model, and fusion model. A movie recommender system has been implemented as an application and the recommendation capability has been evaluated with the cold start problem.