A Deep Architecture for Content-based Recommendations Exploiting Recurrent Neural Networks

Abstract

In this paper we investigate the effectiveness of Recurrent Neural Networks (RNNs) in a top-N content-based recommendation scenario. Specifically, we propose a deep architecture which adopts Long Short Term Memory (LSTM) networks to jointly learn two embeddings representing the items to be recommended as well as the preferences of the user. Next, given such a representation, a logistic regression layer calculates the relevance score of each item for a specific user and we returns the top-N items as recommendations. In the experimental session we evaluated the effectiveness of our approach against several baselines: first, we compared it to other shallow models based on neural networks (as Word2Vec and Doc2Vec), next we evaluated it against state-of-the-art algorithms for collaborative filtering. In both cases, our methodology obtains a significant improvement over all the baselines, thus giving evidence of the effectiveness of deep learning techniques in content-based recommendation scenarios and paving the way for several future research directions.