Topic-enhanced Graph Neural Networks for Extraction-based Explainable Recommendation

Review information has been demonstrated beneficial for the explainable recommendation. It can be treated as training corpora for generation-based methods or knowledge bases for extraction-based models. However, for generation-based methods, the sparsity of user-generated reviews and the high complexity of generative language models lead to a lack of personalization and adaptability. For extraction-based methods, focusing only on relevant attributes makes them invalid in situations where explicit attribute words are absent, limiting the potential of extraction-based models. To this end, in this paper, we focus on the explicit and implicit analysis of review information simultaneously and propose a novel Topic-enhanced Graph Neural Networks (TGNN) to fully explore review information for better explainable recommendations. To be specific, we first use a pre-trained topic model to analyze reviews at the topic level, and design a sentence-enhanced topic graph to model user preference explicitly, where topics are intermediate nodes between users and items. Corresponding sentences serve as edge features. Thus, the requirement of explicit attribute words can be mitigated. Meanwhile, we leverage a review-enhanced rating graph to model user preference implicitly, where reviews are also considered as edge features for fine-grained user-item interaction modeling. Next, user and item representations from two graphs are used for final rating prediction and explanation extraction. Extensive experiments on three real-world datasets demonstrate the superiority of our proposed TGNN with both recommendation accuracy and explanation quality.