Learning Implicit Sentiment for Explainable Review-Based Recommendation

Users can publish reviews to express their detailed feelings about the items. Positive and negative sentiments about various aspects of an item co-existing in the same review may cause confusion in recommendations and generate inappropriate explanations. However, most current explainable recommendation methods fail to capture users' implicit sentiment behind the reviews. In this paper, we propose a novel Implicit Sentiment learning model for Explainable review-based Recommendation, named ISER, which learns users' implicit sentiments from reviews and explores them to generate recommendations with more fine-grained explanations. Specifically, we first propose a novel representation learning to model users/items based on the implicit sentiment behind the reviews. Then we propose two implicit sentiment fusion strategies for rating prediction and explanation generation respectively. Finally, we propose a multi-task learning framework to jointly optimize the rating prediction task and the explanation generation task, which improves the recommendation quality in a mutual promotion manner. The experiments demonstrate the effectiveness and efficiency of our proposed model compared to the baseline models.