A SELF-SUPERVISED OPINION OBJECT CLASSIFICATION OF MICROBLOG’S COMMENTS INVOLVED IN THE CASE BASED ON CONTRASTIVE LEARNING

The classification of opinion objects in Microblog’s comments involved in the case aims to identify the opinion objects in the comments and assign them to categories such as judicial organs, parties, and charges. Aimed at the problem that traditional methods are difficult to extract effective emotional features due to the lack of obvious opinion object words in Microblog’s comments, a self-supervised opinion object classification of Microblog’s comments involved in the case based on contrastive learning is proposed. A multi-head attention global information augmentation module was used to capture key segments in reviews, and the method of contrastive learning was introduced to enhance the text features related to opinion object. The experimental results on the constructed opinion object classification dataset related to Microblog news show that the proposed method improves the macro-average F1 value by 2.2 percentage points compared with the existing benchmark model.