Review Based Recommendations with Human-like Reasons

Recommendation Systems are widely deployed for all kinds of services across various websites to enhance user experience. However, existing systems do not make efficient use of text data associated with products and users, available as reviews and blogs to relate them better. Many recent works have tried to improve the accuracy of rating prediction. However, very few works have attempted to justify the reason for a particular recommendation. Explaining the recommendation would help in gaining the trust of the user, and lend the system human-like credibility. In this paper, we propose a model that can recommend movies and generate a reasoning text to help the user understand why a film was recommended to them. We use three parallel neural networks with an enhanced BERT Embedding for Aspect Based Sentiment Analysis (ABSA) to predict rating. The Seq2Seq transformer model is used to generate the reasoning text.