SEE-3D: Sentiment-driven Emotion-Cause Pair Extraction Based on 3D-CNN

Xin Xu, Guangli Zhu, Houyue Wu, Shunxiang Zhang, Kuan-Ching Li

As an emotional cause detection task, Emotion-Cause Pair Extraction (ECPE) provides technical support for intelligent psychological counseling, empty-nest elderly care, and other fields. Current approaches mainly focus on extracting by recognizing causal relationships between clauses. Different from these existing methods, this paper further considers the influence of sentimental intensity to improve extraction accuracy. To address this issue, we propose an extraction model based on sentiment analysis and 3D Convolutional Neural Networks (3D-CNN), named SEE-3D. First, to prepare fundamental data for sentiment analysis, emotion clauses are clustered into six emotion domains according to six emotion types in the ECPE dataset. Then, a pre-trained sentiment analysis model is introduced to compute emotional similarity, which provides a reference for identifying emotion clauses. In the extraction process, similar features of adjacent documents in the same batch of samples are fused as input of 3D-CNN. The 3D-CNN enhances the macro semantic understanding ability of the model, thereby improving the extraction performance. The results of experiments show that the accuracy of ECPE can be effectively improved by the SEE-3D model.