An Intelligent Recommendation System for TV Programme

Abstract

A personalized recommendation system for multiple users used in TV-Anytime context is presented. The system includes five agents: filtering agent, recommendation agent, profile updating agent, report agent, and interface agent. It also has a user preference database and a database for recommended program information and content. The ``like'' and ``dislike'' information of the users is included in the user preference database. The filtering and recommendation agents propose contents based on the ranking of similarity of user preferences and programme metadata. The user interface agent builds and updates the user profile based on explicit feedback, and collects information on user's reaction to the recommended contents and viewing behavior. This system has sensibility and adaptability to the status of itself and outside and can represent the users' potential needs based on implicit feedback and learn potential changes of their preferences, avoiding the limitation of recommendation based on only explicit needs. Experiment results show the recommendation system can recommend contents effectively.