A Symmetric Local Search Network for Emotion-Cause Pair Extraction

• DOI: 10.18653/v1/2020.coling-main.12
• Publication Date: 2021-01-08T13:58:31Z
• Authors: Zifeng Cheng; Zhiwei Jiang; Yafeng Yin; Hua Yu; Qing Gu
• Keywords: Subnetwork; Causality; Representation

Emotion-cause pair extraction (ECPE) is a new task which aims at extracting the potential clause pairs of emotions and corresponding causes in document. To tackle this task, two-step method was proposed by previous study first extracted emotion clauses cause individually, then paired clauses, filtered out without causality. Different from that separated detection matching into two steps, we propose Symmetric Local Search Network (SLSN) model to perform simultaneously local search. SLSN consists symmetric subnetworks, namely subnetwork subnetwork. Each composed representation learner searcher. The searcher specially-designed cross-subnetwork component can extract emotion-cause pairs. Experimental results on ECPE corpus demonstrate superiority our over existing state-of-the-art methods.