Effective Inter-Clause Modeling for End-to-End Emotion-Cause Pair Extraction

Abstract

Emotion-cause pair extraction aims to extract all emotion clauses coupled with their cause clauses from a given document. Previous work employs two-step approaches, in which the first step extracts emotion clauses and cause clauses separately, and the second step trains a classifier to filter out negative pairs. However, such pipeline-style system for emotion-cause pair extraction is suboptimal because it suffers from error propagation and the two steps may not adapt to each other well. In this paper, we tackle emotion-cause pair extraction from a ranking perspective, i.e., ranking clause pair candidates in a document, and propose a onestep neural approach which emphasizes interclause modeling to perform end-to-end extraction. It models the interrelations between the clauses in a document to learn clause representations with graph attention, and enhances clause pair representations with kernel-based relative position embedding for effective ranking. Experimental results show that our approach significantly outperforms the current two-step systems, especially in the condition of extracting multiple pairs in one document.

Multi-Task Learning and Adapted Knowledge Models for Emotion-Cause Extraction

“…However, comparatively few researchers have looked at the semantic roles related to emotion such as the cause, the target or the experiencer, with few exceptions for Chinese (Gui et al, 2016;Chen et al, 2018;Wei et al, 2020;Ding et al, 2020), English Ghazi et al, 2015;Kim and Klinger, 2018;Bostan et al, 2020; and Italian (Russo et al, 2011). We highlight some of these works here and draw connection to our work.…”

A Mutually Auxiliary Multitask Model With Self-Distillation for Emotion-Cause Pair Extraction

“…Wu et al [41] solved the relationship classification task together with emotion and cause extraction in a unified model. Wei et al [42] proposed a one-step approach to emphasize inter-clause modeling from a ranking perspective. Ding et al [43] designed a 2D Transformer and its two variants to model the interaction between emotion-cause pairs.…”

A Unified Sequence Labeling Model for Emotion Cause Pair Extraction

“…• RANKCP: proposed by Wei et al (2020), they made use of the graph attention network to propagates information among clauses and ranked the candidate ECPs with the learned pair representations to get the prediction results.…”

“…Recently, there have been several works focusing on the ECPE task, aiming to propose end-to-end approaches to avoid the possible cascading errors brought by the two-step manner in the three approaches proposed by . Wei et al (2020) proposed a RANKCP model which makes use of the graph attention network to propagates information among clauses and ranks the candidate ECPs based on the learned pair representations to get the prediction results. Tang et al (2020) proposed a BERTbase model called LAE-Joint-MANN-BERT to deal with the emotion detection (ED) and the ECPE task jointly.…”