An important task for recommender system is to generate explanationsaccording to a user's preferences. Most of the current methods for explainablerecommendations use structured sentences to provide descriptions along with therecommendations they produce. However, those methods have neglected thereview-oriented way of writing a text, even though it is known that thesereviews have a strong influence over user's decision. In this paper, we propose a method for the automatic generation of naturallanguage explanations, for predicting how a user would write about an item,based on user ratings from different items' features. We design acharacter-level recurrent neural network (RNN) model, which generates an item'sreview explanations using long-short term memories (LSTM). The model generatestext reviews given a combination of the review and ratings score that expressopinions about different factors or aspects of an item. Our network is trainedon a sub-sample from the large real-world dataset BeerAdvocate. Our empiricalevaluation using natural language processing metrics shows the generated text'squality is close to a real user written review, identifying negation,misspellings, and domain specific vocabulary.
展开▼
