Exploring Auxiliary Information Integration for Neural TV Program Recommendation

TV program recommendations enable users to discover programs of interest within the large amount of information available. Most of the existing methods use userprogram interaction data directly based on collaborative filtering methods and do not exploit the potential value of auxiliary information. Furthermore, neural networks and attention mechanisms can effectively integrate different types of auxiliary information to achieve accurate representations of users and programs. We propose a neural TV program recommendation based on auxiliary information (NPR-AI) and discuss its effectiveness. In the program representation, we perform contextsensitive and context-free information encoders according to different types of auxiliary information, and then use neural networks and multi-layer attention to identify multi-hierarchical program information. In the user representation, we examine the programs a user has viewed and employ a personalized attention mechanism to determine the significance of each program. We demonstrate in experiments on real datasets that our approach greatly improves the effectiveness of TV program recommendations.