The goal of Emotion Cause Pair Extraction task is to automatically extract emotional reasons and their corresponding relationships from text.Therefore,which is more challenging than Emotion Cause Extraction.The existing methods for Emotion Cause Pair Extraction in text often overlook the bidirectional dependency between emotions and causes.To address this issue,we propose decomposing Emotion Cause Pair Extraction task into sub-sentence level sequence labeling and label pairing tasks.Firstly,we use pre-trained language models,bidirectional long short-term memory networks,and attention mechanisms to obtain sub-sentence vectors.Then,we adopt convolutional neural networks to capture adjacent information between sentences,and combine it with a conditional random field for sequence labeling.Finally,we use rule-based methods to complete the emotion-cause pairing.The experimental results show that the precision for emotional clause prediction is 0.726,the recall is 0.688,and the Fl score is 0.703 for the emotion-cause pair matching task,thus validating the effectiveness of the proposed method.
sentiment-reason pairingdeep learningpre-trained language modelsequence labeling
NETL
NSTL
万方数据
