A News Recommendation Model Based on Time Awareness and News Relevance

Personalized news recommendation can target user interests and effectively alleviate information overload. Most of the existing methods are based on news content for recommendation, which mostly ignore the rich auxiliary information and neighbor information existing in real news recommendation scenarios. In addition, few methods provide easy-to-understand explanations. In this paper, we propose a news recommendation model based on time awareness and news relevance. The model combines various news auxiliary information and user-news interaction data in the form of heterogeneous graph, and mines the temporal relationship in the user click sequence for news recommendation. In addition, our model provide understandable recommendation explanations based on the multiple explanation bases extracted from the heterogeneous graph. Extensive experiments on two public and widely used datasets, Adressa and Globo, demonstrate both the effectiveness of the proposed approach and the reasonableness of recommendation explanations.