Multi-Task Learning Network for Document-level and Multi-aspect Sentiment Classification
Zhou Wang, Jing Cao · 2020
Document-level sentiment classification aims to predict overall sentiment polarity in a document about a product, while multi-aspect sentiment classification aims at detecting sentiment polarities for different aspects of a product in a document. Most existing methods perform the two tasks separately and ignore the correlation between them. In this paper, we propose a multi-task framework called multi-sentiment hierarchical attention network (MSHAN) that jointly performs document-level and multi-aspect sentiment classification both. Specifically, MSHAN adopts hierarchical architecture and attention mechanism to predict aspect sentiments and aggregate those aspect sentiments into overall sentiment. Moreover, considering that aspect sentiment can not fully express overall sentiment, MSHAN adopts another hierarchical architecture to capture additional document sentiment information and add this sentiment to the overall sentiment. Experimental results on two real-world datasets show that the proposed method outperforms previous methods. To the best of our knowledge, this is the first study that performs document-level and multi-aspect sentiment classification in a unified model.