[4Xin2] Poster session 2

[4Xin2-04] Stance Detection Using Contrastive Learning and External Knowlegde

〇Tomoki Yoshino¹, Minoru Sasaki¹ (1.Ibaraki University)

Keywords:Stance Detection

In zero-shot stance detection, learning topic-invariant expressions and using external knowledge have been highly successful. In these methods, the PT-HCL model presents an effective approach to zero-shot stance detection using contrastive learning. In this study, we propose a novel stance detection model that is trained simultaneously by adding external knowledge to a topic-invariant expressions. As a result, we found an improvement in accuracy for four topics on the Sem16 dataset compared to the case where Wikipedia is not used. However, the learning speed slowed down due to the increase in the amount of information. There are still many issues to be addressed, and we expect to improve the detection accuracy by revising the evaluation method in the training phase and adjusting the weights of the training data and Wikipedia information.