Learning to Rank Tweets with Author-Based Long Short-Term Memory Networks

Recommending tweets that a user might retweet plays an important role either in satisfying users' information needs or in the dissemination of information in microblogging services such as Twitter. In this paper, we propose a deep neural network for tweet recommendations with author-based Long Short-Term Memory networks for learning the latent representations/embeddings of tweets. Our approach predicts the preference score of a tweet based on (1) the similarity between the embeddings of a user and the tweet, (2) the similarity between the embeddings of the user and the author (who posted the tweet). Despite its simplicity, we present that our approach can significantly outperform state-of-theart methods with or without explicit features for recommending tweets in terms of five evaluation metrics.