Fin-MLGAT: Line Graph Attention Network for Dialogue Emotion Recognition in Stock Market and its Evaluations

Xiaobing Feng, Ling Feng, Xiujuan Xu, Jiaqi Qiao, Wenjian Zhang

Abstract: With the rapid development of social media and digitalization, the task of emotion pattern recognition and extreme emotion detection in conversation is playing an increasingly important role in understanding emotion interaction among and between parties. Traditional methods employ probabilistic statistical models that leverage prior knowledge to construct models However when dealing with complex patterns, large scale data, and high-dimensional features, their performance becomes limited. Alternatively, deep learning methods are capable of automatically learning highly abstract feature representations from raw datasets. Our Fin-MLGAT mines the emotional information dependent on the target utterance by focusing on local con text and global context at different scales. Experiments show that our model achieves the second-highest performance among all current methods by Wa-F1. Moreover, the model can show high accuracy in a few classes without additional measures.