Online Opinion Conflict Interaction Recognition Based on Dependent Multi-Task Deep Learning

Abstract

Expression of conflicting opinions often appear in the interaction process of different users in various online communities, and effectively recognizing opinion conflict interactions is of great significance. Compared to traditional online opinion mining tasks, such as stance/sentiment classification, the opinion conflict interaction recognition shows unique and challenging characteristics that it is determined based on the interaction between the differences in emotion expression and the consistency of thematic content given two opinions. Thus, the paper tries to propose a model that can effectively model its unique feature and recognize different interaction categories, that is, Opinion Conflict Interaction Recognition based on Causal Inference-enhanced dependent Multi-Task Deep Learning, noted as CIMDL. The model introduced casual inference-enhanced multi-task learning and co-attention interaction mechanism in addition to the pre-training language model-based text embedding and deep neural network-based feature extraction. We construct two benchmark datasets for the new proposed task, and conduct extensive experiments to demonstrate the advantages of the proposed approach over different state-of-arts baselines.