Graph Convolutional Networks for Aspect-Based Sentiment Analysis

Abstract

Aspect-based sentiment analysis, as an important fine-grained sentiment analysis problem, aims to analyze and understand the emotions at the aspect level in sentences. However, existing models often overlook the syntactic relationship between words and fail to extract specific semantic information.We propose a graph convolutional network model to extract aspectual word features from local context. This paper also proposes a model to address the problem of under-use of explicit syntactic dependency in aspect category emotion analysis, based on a graph convolutional network and incorporating external knowledge to extract deep and surface structure information from the dependency graph, using the aspect item as a reference point. Both models are superior to existing models.