Discrete Opinion Tree Induction for Aspect-based Sentiment Analysis

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 2051 - 2064 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Discrete Opinion Tree Induction for Aspect-based Sentiment Analysis Chenhua Chen1,2, Zhiyang Teng1,2, Zhongqing Wang3 and Yue Zhang1,2 1School of Engineering, Westlake University, China 2Institute of Advanced Technology, Westlake Institute for Advanced Study, 3Soochow University, {chenchenhua, tengzhiyang}@westlake.edu.cn, wangzq@suda.edu.cn, yue.zhang@wias.org.cn Abstract Dependency trees have been intensively used with graph neural networks for aspect-based sentiment classification. Though being effec- tive, such methods rely on external depen- dency parsers, which can be unavailable for low-resource languages or perform worse in low-resource domains. In addition, dependency trees are also not optimized for aspect-based sentiment classification. In this paper, we pro- pose an aspect-specific and language-agnostic discrete latent opinion tree model as an alter- native structure to explicit dependency trees. To ease the learning of complicated structured latent variables, we build a connection between aspect-to-context attention scores and syntac- tic distances, inducing trees from the attention scores. Results on six English benchmarks, one Chinese dataset and one Korean dataset show that our model can achieve competitive perfor- mance and interpretability. 1 Introduction Aspect-based sentiment classification (ABSA) is the task of recognizing the sentiment polarities of specific aspect categories or aspect terms in a given sentence (Jiang et al., 2011; Dong et al., 2014; Wang et al., 2016; Tang et al., 2016; Li et al., 2018; Du et al., 2019; Sun et al., 2019a; Seoh et al., 2021; Xiao et al., 2021). Different from document-level sentiment analysis, different aspect terms in the same document can bear different sentiment polari- ties. For example, given a restaurant review “decor is nice though service can be spotty", the corre- sponding sentiment labels of “decor” and “service” are positive and negative, respectively. How to locate the corresponding opinion con- texts for each aspect term is a key challenge for ABSA. To this end, recent efforts leverage depen- dency trees (Zhang et al., 2019; Sun et al., 2019a; Wang et al., 2020). Syntactic dependencies have been shown to better capture the interaction be- tween the aspect and the opinion contexts (Huang decor is be nice though service can spotty (a) dependency tree. (b) Induced tree for “decor”. (c) Induced tree for “service”. Figure 1: A dependency tree of the input sentence “decor is nice though service can be spotty” and two induced trees of two aspects in this sentence. et al., 2020; Tang et al., 2020). For example, in Figure1(a), using syntactic relations, we can find that the corresponding opinion words for “decor” and “service” are “nice” and “spotty“, respectively. Despite its effectiveness, dependency syntax has the following limitations. First, dependency parsers can be unavailable for low-resource languages or perform worse in low-resource domains (Duong et al., 2015; Rotman and Reichart, 2019; Vania et al., 2019; Kurniawan et al., 2021). Second, de- pendency trees are also not optimized for aspect- based sentiment classification. Previous stud- ies transform dependency trees to aspect-specific forms by hand-crafted …

Content truncated to 500 words