Aspect-Specific Heterogeneous Graph Convolutional Network for Aspect-Based Sentiment Classification

TL;DR: This work proposes a novel GCN-based model that uses a heterogeneous graph to identify the sentiment expressed towards an aspect given a context sentence and shows that the network consistently outperforms the state-of-the-art model on all these datasets.

Abstract: Aspect-based sentiment classification aims to identify the sentiment expressed towards an aspect given a context sentence. There are two main problems with existing methods: First, the methods simply take the average of the sentence and aspect word vectors as the sentence and aspect representations for a certain sentence, but they are not explicit representations and will lose considerable useful information. Second, existing models based on graph convolutional networks (GCNs) only use the dependency relationship of a sentence, which cannot fully exploit the potential of the sentence and exert the powerful feature fusion ability of GCNs. To solve these problems, we propose a novel GCN-based model that uses a heterogeneous graph. We explicitly define sentence and aspect nodes to learn the sentence and aspect representations separately and then combine 4 kinds of relationships to construct the heterogeneous graph. In our experiments conducted on 5 public datasets, the experimental results show that our network consistently outperforms the state-of-the-art model on all these datasets.