Making Flexible Use of Subtasks: A Multiplex Interaction Network for Unified Aspect-based Sentiment Analysis

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 2695–2705 August 1–6, 2021. ©2021 Association for Computational Linguistics 2695 Making Flexible Use of Sub-tasks: A Multiplex Interaction Network for Unified Aspect-based Sentiment Analysis Guoxin Yu♠♥, Xiang Ao♠♥∗, Ling Luo♠♥, Min Yang♣, Xiaofei Sun♦, Jiwei Li♦, Qing He♠♥ ♠Key Lab of Intelligent Information Processing of Chinese Academy of Sciences (CAS), Institute of Computing Technology, CAS, Beijing 100190, China. ♥University of Chinese Academy of Sciences, Beijing 100049, China. ♣Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences ♦Shannon.AI, China. {yuguoxin20g,aoxiang,luoling18s,heqing}@ict.ac.cn min.yang@siat.ac.cn {xiaofei sun,jiwei li}@shannonai.com Abstract Aspect Term Extraction (ATE), Opinion Term Extraction (OTE) and Aspect Sentiment Clas- sification (ASC) are the essential building blocks of Aspect-based Sentiment Analy- sis (ABSA). They are typically treated as sep- arate tasks and are individually studied by previous work. Recent studies intend to in- corporate multiple sub-tasks into a unified framework, but suffer from the following ma- jor disadvantages: (1) ABSA models are ex- tremely fragile when some sub-tasks are ab- sent; (2) the interactive relations among sub- tasks are not adequate. To this end, we pro- pose a multi-task learning approach named MIN (Multiplex Interaction Network) to make flexible use of sub-tasks for a unified ABSA. We divide the sub-tasks of ABSA into ex- tractive sub-tasks and classification sub-tasks, and optimize these sub-tasks in a unified man- ner with multiplex interaction mechanisms. Specifically, we devise a pairwise attention to exploit bidirectional interactions between any arbitrary pair of extractive sub-tasks and a consistency-weighting to perform unidirec- tional interaction from an extractive sub-task to a classification sub-task. Since the proposed interaction mechanisms are task-agnostic, our model can also work well when some specific sub-tasks are absent. Extensive experiments on two widely used benchmarks with different numbers of sub-tasks demonstrate the superi- ority of the proposed model. 1 Introduction Aspect-based sentiment analysis (ABSA), a fine- grained task of text sentiment analysis (Liu, 2012), ∗Corresponding author aims at summarizing opinions in user comments towards different targets (also known as aspects). ABSA generally consists of three specific sub-tasks, namely, aspect term extraction (ATE), opinion term extraction (OTE) and aspect sentiment classifica- tion (ASC). For example, consider the sentence “The price is reasonable although the service is in- ferior.”, ATE aims to extract a set of aspect terms from the sentence, i.e. {price, service}, OTE ex- tracts the opinion words, i.e. {reasonable, inferior} , and ASC predicts sentiment polarity for each as- pect that is positive over the first aspect price and negative for the second aspect service, respectively. Prevailing solutions of ABSA treated ATE (Liu et al., 2015; Li and Lam, 2017; Angelidis and Lap- ata, 2018; Liao et al., 2019; Luo et al., 2019b; Ma et al., 2019), OTE (Wang et al., 2017; Wang and Pan, 2019) and ASC (Wang et al., 2016b; Chen et al., 2017; He et al., 2018; Li et al., 2018b; Du et al., 2019; Xu et al., 2021) as separate tasks and were individually studied for decades. These sepa- rate tasks need to …

Content truncated to 500 words