A graph attention network utilizing multi-granular information for emotion-cause pair extraction

Emotion-cause pair extraction (ECPE) aims to extract emotion and cause clauses underlying a text and pair them. Most of the recent approaches to this problem adopt deep neural networks to model the inter-clause dependency, without making full use of information at word level and document level. In this paper, we propose a model that utilizes multi-granular information, including word-level, clause-level, and document-level information, to facilitate emotion-cause pair extraction. Our model consists of two fully-connected clause graphs, including emotion graph and cause graph, and graph attention is applied to learn emotion-specific and cause-specific representations which are then used to generate document-level representations. To exploit the mutual indication between emotion and cause, a cross-graph co-attention mechanism is proposed. Moreover, external knowledge of emotional and causal cues is incorporated to provide word-level indicative information for emotion-cause pair extraction. The proposed model is tested on both Chinese [1] and English [2] datasets, and the results show that our model achieves the state-of-the-art performance on both datasets