TRRS: Temporal Recurrent Recommender System based on Time-sync Comments

Recent years has witnessed great emerge of online video websites, including the exploded number of videos and users. As a result, there appears a lot of personlized recommender systems. However there remain some challenging problems to tackle such as cold start problem, which scientists have made use of all kinds of sideinformation, e.g. gender, age or comments, to release. Currently a new type of video comments, called TSCs (TSC), plays a more and more important role in video watching activity. In this paper we utilize TSC to recommend videos for users. We developed a deep nueral network model called Temporal Recurrent Recommder System (TRRS) which combine multi-layers neural network to extract feature for users and videos. The first layer convert TSC to embeddings, then RNN layer analyze each comment from user or video, and fianlly the merge layer combine all output from prior layer and produce the feature. We use the feature from the network for users and videos to make personlized recommendation.