Case-based Reasoning for Resolving the Diversity-accuracy Dilemma of Recommendation System through Subdivision

Recommendation systems are widely used to help their netizens find what they want from cyber space and many various applications have been developed in different areas like news feeding, entree recommending and online shopping, etc. Many researches have found that their system performance is obsession with diversity-accuracy dilemma, and recommender has to balance recommendation sequence results. CBR-recommender is suggested for it has a comprehensive expression of human sense, logics and result explanation, and has more system flexibility to integrate many AI tools in dealing with such diversity-accuracy dilemma. Covering algorithm is also proposed to dynamically cluster similar users or items as subdomain for which all users are dynamitic divided into multiple clusters according to a certain criterion. Our experiments results indicate that users with similar hobbies are assigned to the same sub-domain as many specific sub-classes, and the refined classification results are more conducive for the problem-solving.