A Multi-level Network with Multi-feature Clause Pair Graph for Emotion Cause Pair Extraction

Kai Kang,Guozheng Rao,Li Zhang,Qing Cong,Xin Wang

DOI: https://doi.org/10.1007/978-981-97-2303-4_30

2024-01-01

Abstract:Emotion-cause pair extraction (ECPE) is an emerging task which aims to extract all emotion clauses and corresponding cause clauses in a document. Most current methods use an end-to-end framework with two auxiliary tasks: emotion extraction and cause extraction, to solve this problem. However, these methods ignore the inter-task interaction among the three tasks and independently extract the features of emotion, cause, and emotion-cause pair. At the same time, most approaches do not fully utilize phrase-level information that is important for the ECPE task. To address these problems, we propose a novel multi-level network with multi-feature clause pair graph to extract different information from the document and generate multi-task features simultaneously. We use word-level and phrase-level network to capture the word and phrase information. At the clause-level, we construct a multi-feature clause pair graph to model the relationship between emotion, cause, and pair. Experimental results show that our method outperforms other baselines on a standard ECPE corpus.