Aspect-Based Sentiment Analysis with Multi-aspects Heterogeneous Graph Convolutional Networks

Abstract

The target of aspect-based sentiment analysis (ABSA) mission is in order to perform emotional polarity judgments of the specified aspect among one data set, that is, a sentence. The current models ignore the interaction of multiple aspects within a sentence, and the representation of aspect and context information is inadequate. To solve these problems, we integrate multi-aspects and contextual information into a graph, then put forward a multi-aspects heterogeneous graph convolutional network (MAHGCN) model to update and represent nodes. It is verified by experiments on four data sets that MAHGCN model achieves significant and consistent improvement as compared to other baselines.