Utilizing Content to Enhance a Usage-Based Method for Web Recommendation based on Q-Learning

The problem of information overload on the Internet has received a great deal of attention in the recent years. Recommender Systems have been introduced as one solution to this problem. These systems aim at directing the user toward the items that best meet her needs and interests. Recent studies have indicated the effectiveness of incorporating domain knowledge in improving the quality of recommendations. In this paper we exploit this approach to enhance a reinforcement learning framework, primarily devised for web recommendations based on web usage data. A hybrid, i.e. content- and usage-based, web recommendation method is proposed by incorporating web content information into a model of user behavior learned form usage data. Content information is utilized to find similarities between usage scenarios, i.e. users ' seeking their information needs, and new recommendation strategies are proposed that are based on this enhanced model of user behavior. We evaluate our method under different settings and show how this method can overcome the shortcomings of the usage-based approach and improve the overall quality of recommendations