Fusing Syntax and Semantics-Based Graph Convolutional Network for Aspect-Based Sentiment Analysis

Abstract

Aspect-based sentiment analysis (ABSA) aims to classify the sentiment polarity of a given aspect in a sentence or document, which is a fine-grained task of natural language processing. Recent ABSA methods mainly focus on exploiting the syntactic information, the semantic information and both. Research on cognition theory reveals that the syntax an*/874d the semantics have effects on each other. In this work, a graph convolutional network-based model that fuses the syntactic information and semantic information in line with the cognitive practice is proposed. To start with, the GCN is taken to extract syntactic information on the syntax dependency tree. Then, the semantic graph is constructed via a multi-head self-attention mechanism and encoded by GCN. Furthermore, a parameter-sharing GCN is developed to capture the common information between the semantics and the syntax. Experiments conducted on three benchmark datasets (Laptop14, Restaurant14 and Twitter) validate that the proposed model achieves compelling performance comparing with the state-of-the-art models.