Automatic Generation of Natural Language Explanations

An interesting challenge for explainable recommender systems is to provide successful interpretation of recommendations using structured sentences. It is well known that user-generated reviews, have strong influence on the users’ decision. Recent techniques exploit user reviews to generate natural language explanations. In this paper, we propose a character-level attention-enhanced long short-term memory model to generate natural language explanations. We empirically evaluated this network using two real-world review datasets. The generated text present readable and similar to a real user’s writing, due to the ability of reproducing negation, misspellings, and domain-specific vocabulary.