DeepLTRS: A deep latent recommender system based on user ratings and reviews

We introduce a deep latent recommender system named deepLTRS in order to provide users with high quality recommendations based on observed user ratings and texts of product reviews. The underlying motivation is that, when scores only few products, the used reviews represent significant source information, thereby enhancing predictive ability model. Our approach adopts variational auto-encoder (VAE) architecture as generative model for an ordinal matrix encoding document-term Taking into account both matrices inputs, uses neural network capture relationship between factors topics. Moreover, user-majoring encoder product-majoring are constructed jointly preferences. Due specificity structure, original row-column alternated mini-batch optimization algorithm proposed deal user-product dependencies computational burden. Numerical experiments simulated real-world data sets demonstrate that outperforms state-of-the-art, particular context extreme sparsity.