SSEGCN: Syntactic and Semantic Enhanced Graph Convolutional Network for Aspect-based Sentiment Analysis

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4916 - 4925 July 10-15, 2022 ©2022 Association for Computational Linguistics SSEGCN: Syntactic and Semantic Enhanced Graph Convolutional Network for Aspect-based Sentiment Analysis Zheng Zhang, Zili Zhou∗, Yanna Wang School of Cyber Science and Engineering, Qufu Normal University, China zhangzhengzz12@gmail.com, zzl@qfnu.edu.cn, wyn@qfnu.edu.cn Abstract Aspect-based Sentiment Analysis (ABSA) aims to predict the sentiment polarity towards a particular aspect in a sentence. Recently, graph neural networks based on dependency tree con- vey rich structural information which is proven to be utility for ABSA. However, how to ef- fectively harness the semantic and syntactic structure information from the dependency tree remains a challenging research question. In this paper, we propose a novel Syntactic and Seman- tic Enhanced Graph Convolutional Network (SSEGCN) model for ABSA task. Specifically, we propose an aspect-aware attention mech- anism combined with self-attention to obtain attention score matrices of a sentence, which can not only learn the aspect-related semantic correlations, but also learn the global semantics of the sentence. In order to obtain comprehen- sive syntactic structure information, we con- struct syntactic mask matrices of the sentence according to the different syntactic distances between words. Furthermore, to combine syn- tactic structure and semantic information, we equip the attention score matrices by syntactic mask matrices. Finally, we enhance the node representations with graph convolutional net- work over attention score matrices for ABSA. Experimental results on benchmark datasets il- lustrate that our proposed model outperforms state-of-the-art methods. 1 Introduction Aspect-based Sentiment Analysis (ABSA) aims to determine the sentiment polarity of a given as- pect term in a sentence, where sentiment polarity includes positive, negative and neutral. For exam- ple, in Figure 1, ABSA determines the sentiment towards the aspects “food” and “service”. For aspect term “food”, the sentiment polarity is neg- ative, but “service” is positive. That is, we need to discriminate sentiment polarities according to ∗Corresponding author. Figure 1: An example sentence with its dependency tree where aspect terms (highlighted in red) are connected with other words based on their syntactic dependencies. different aspects. The main idea of most works is to model the dependency relation between aspects and their associated opinion words. Prior studies exploit attention mechanism (Wang et al., 2016; Chen et al., 2017; Ma et al., 2017; Liu et al., 2018; Hu et al., 2018; Wang et al., 2018; Huang et al., 2018; Fan et al., 2018) to model the correlations between aspect term and the con- text. However, attention mechanism is vulnerable to noise in sentences, i.e., the irrelevant words. Recent works on ABSA leverage Graph Neural Networks (GNNs) over the dependency tree of a sentence to exploit syntactic structure (Sun et al., 2019; Zhang et al., 2019; Liang et al., 2020; Wang et al., 2020; Zhao et al., 2020; Li et al., 2021; Tian et al., 2021). In these works, Zhang et al. (2019) employ Graph Convolutional Network (GCN) to in- tegrate the syntactic information. Sun …

Content truncated to 500 words