PTMT: Multi-Target Stance Detection with PTM-enhanced Multi-Task Learning

Abstract

Stance detection, as a sub-task of sentiment analysis, is becoming an essential tool in the field of online public opinion analysis with the rapid development of social media. At the moment, the overwhelming majority of stance detection methods are only focused on a single target. However, in an electoral scene, those single-target methods may lose some interrelated information in multi-target sentences. This paper proposes a pre-trained model based multi-target stance detection approach to automatically dig out the implicit targets' interrelated information and words of interest via the attention mechanism under the multi-task learning framework. And to involve the hashtags' semantic information, some specific dataset preprocessing methods are also applied here. By comparing various methods, we show that our model achieves state-of-the-art results in a benchmark dataset.