AN AGENT-BASED RECOMMENDER SYSTEM USING IMPLICIT FEEDBACK, IMPROVED INTER-USER SIMILARITY AND RATING PREDICTION

This paper proposes the architecture of an agent-based hybrid recommender system consists of three subsystems: content-based, collaborate filtering and rating prediction. They are controlled by a task agent, who self-determines when to execute the hybrid recommendation algorithm based on user’s implicit feedback and its knowledge base. In content-based subsystem, an improved implicit feedback algorithm is presented. Moreover, an improved inter-user similarity function is applied to find out like-minded neighbors in collaborate filtering subsystem. To enhance the quality, result sets of content-based and collaborate filtering recommendations are reunited to select the top 5 items through our rating prediction algorithm based on linear regression model. Afterwards, a client agent sends back recommendations and receives another user’s request. Experimental data show that hybrid techniques hold the promise of produce high-quality recommendations with the assistance of agents.