A Deep Joint Network for Session-based News Recommendations with Contextual Augmentation.

Session-based recommendations have drawn more and more attention in many recommendation settings of modern online services. Unlike many other domains such as books and music, news recommendations suffer from new challenges of fast updating rate and recency issues of news articles and lack of user profiles. In this paper, we proposed a method that combines user click events within session and news contextual features to predict next click behavior of a user. The model consists of two different kinds of hierarchical neutral networks to learn article contextual properties and temporal sequential patterns in streams of clicks. Character-level embedding over input features is adopted to allow integrating different types of data and reduce engineering computation. Besides, we also introduced a time-decay method to compute the freshness of news articles within a time slide. Experimental results on two real-world datasets show significant improvements over several baselines and state-of-the-art methods on session-based neural networks.