Sentiment-Aware Neural Recommendation with Opinion-Based Explanations

Abstract

Explainable recommendation systems are crucial for complex decision making, which provide users with the recommendation results as well as the reasons why such items are recommended. However, most existing explainable recommendation methods only consider one aspect of user sentiment, such as ratings or reviews, which fails to capture the fine-grained user sentiment. In this paper, we propose a novel sentiment-aware neural recommendation model, named SNROE, which jointly performs a rating prediction task and an explanation generation task, to guarantee both the accuracy of recommendation and the personalization of explanations. For the rating prediction task, we adopt MLP to learn user/item representations. For the explanation generation task, we propose a sentiment-aware explanation generation method, which utilizes pretrained Transformer to generate opinion-based explanations by fusing users’ rating-level sentiment, aspect-level sentiment and review-level sentiment. We also propose a joint training algorithm to jointly optimize the above two tasks. The experiments demonstrate the effectiveness and the efficiency of our proposed model compared to the baseline models.

This work was supported by the National Natural Science Foundation of China under Grant Nos. 62072084, 62172082 and 62072086, the Science Research Funds of Liaoning Province of China under Grant No. LJKZ0094, the Natural Science Foundation of Liaoning Province of China under Grant No. 2022-MS-171, the Science and Technology Program Major Project of Liaoning Province of China under Grant No. 2022JH1/10400009.