Pair-Based Joint Encoding with Relational Graph Convolutional Networks for Emotion-Cause Pair Extraction

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 5339 - 5351 December 7-11, 2022 ©2022 Association for Computational Linguistics Pair-Based Joint Encoding with Relational Graph Convolutional Networks for Emotion-Cause Pair Extraction Junlong Liu, Xichen Shang, Qianli Ma∗ School of Computer Science and Engineering, South China University of Technology, Guangzhou, China junlongliucs@foxmail.com qianlima@scut.edu.cn∗ Abstract Emotion-cause pair extraction (ECPE) aims to extract emotion clauses and corresponding cause clauses, which have recently received growing attention. Previous methods sequen- tially encode features with a specified order. They first encode the emotion and cause fea- tures for clause extraction and then combine them for pair extraction. This lead to an imbal- ance in inter-task feature interaction where fea- tures extracted later have no direct contact with the former. To address this issue, we propose a novel Pair-Based Joint Encoding (PBJE) net- work, which generates pairs and clauses fea- tures simultaneously in a joint feature encod- ing manner to model the causal relationship in clauses. PBJE can balance the information flow among emotion clauses, cause clauses and pairs. From a multi-relational perspective, we construct a heterogeneous undirected graph and apply the Relational Graph Convolutional Net- work (RGCN) to capture the various relation- ship between clauses and the relationship be- tween pairs and clauses. Experimental results show that PBJE achieves state-of-the-art per- formance on the Chinese benchmark corpus.1 1 Introduction Emotion cause extraction (ECE) is a kind of emo- tion analysis task which is first proposed by Lee et al. (2010) and has developed for a long time (Gui et al., 2017; Li et al., 2018b, 2019; Hu et al., 2021). ECE extracts the cause for the input document and certain emotion labels. However, emotions in the documents need to be annotated in advance, which requires manual input and takes lots of time (Xia and Ding, 2019). Hence, Xia and Ding (2019) pro- poses a new task called emotion-cause pair extrac- tion (ECPE). Given a document as the input, ECPE extracts the clauses which express emotions and *Corresponding author 1Our codes are publicly available at https://github. com/tutuDoki/PBJE-ECPE C1: The next day C2: The couples talked face to face again for a long time … C5: They quarreled because of the trivial matters again C6: The girl asked to break up in a fit of anger … C10: The girl faced the boy sitting on the bench C11: She couldn't hold back her anger C12: After thinking of the boy's bad treatment of her C13: She took out the simulated toy gun which she carried from her trouser pocket angrily … (C1, C1) ❌ (C1, C2) ❌ � (C6, C5) � (C13, C12) � � Clauses Pairs (Ci, Cj) (C6, C12) ❌ � (C11, C10) ❌ � RankCP Confidience 0.0 0.0 0.002 0.001 0.918 0.605 Figure 1: An example document from the ECPE corpus where ci represents the emotion clause and cj represents the cause clause in pair. The words in red are the key- words about emotion and the words in blue …

Content truncated to 500 words