End-to-End Emotion-Cause Pair Extraction based on Sliding Window Multi-Label Learning

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 3574–3583, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 3574 End-to-End Emotion-Cause Pair Extraction based on Sliding Window Multi-Label Learning Zixiang Ding, Rui Xia∗, Jianfei Yu School of Computer Science and Engineering, Nanjing University of Science and Technology, China {dingzixiang,rxia,jfyu}@njust.edu.cn Abstract Emotion-cause pair extraction (ECPE) is a new task that aims to extract the potential pairs of emotions and their corresponding causes in a document. The existing methods first perfor- m emotion extraction and cause extraction in- dependently, and then perform emotion-cause pairing and filtering. However, the above methods ignore the fact that the cause and the emotion it triggers are inseparable, and the extraction of the cause without specify- ing the emotion is pathological, which great- ly limits the performance of the above meth- ods in the first step. To tackle these short- comings, we propose two joint frameworks for ECPE: 1) multi-label learning for the ex- traction of the cause clauses corresponding to the specified emotion clause (CMLL) and 2) multi-label learning for the extraction of the e- motion clauses corresponding to the specified cause clause (EMLL). The window of multi- label learning is centered on the specified emo- tion clause or cause clause and slides as their positions move. Finally, CMLL and EMLL are integrated to obtain the final result. We evaluate our model on a benchmark emotion cause corpus, the results show that our ap- proach achieves the best performance among all compared systems on the ECPE task. 1 Introduction With the rapid growth of social media, emotion analysis of online text has received much attention in recent years. Unlike most studies that focused on the detection and classification of emotions, (Lee et al., 2010) first proposed the emotion cause extrac- tion (ECE) task, which aims to extract the stimulus behind emotions. (Gui et al., 2016a) released a public corpus and defined the ECE task as a fine- grained emotion analysis task, where the goal is to judge for each clause in the document whether ∗Corresponding author c1: Yesterday morning c2: a policeman visited the old man with the lost money, c3: and told him that the thief was caught. c4: The old man was very happy. c5: But he still feels worried, c6: as he doesn’t know how to keep so much money {c4-c2, c4-c3, c5-c6} Output: emotion-cause pairs Input: a document Figure 1: An example showing the emotion-cause pair extraction (ECPE) task. it is the corresponding cause, given the annotation of emotions. This corpus has received a lot of at- tention in subsequent research and has become a benchmark dataset for the ECE task (Gui et al., 2017; Li et al., 2018; Yu et al., 2019; Xu et al., 2019; Ding et al., 2019; Xia et al., 2019). However, there are several inherent shortcomings in the set- ting of the ECE task: firstly, the need for emotion annotation greatly limits the practical applications of the ECE task; secondly, the way of …

Content truncated to 500 words