Stance Detection on Microblog Focusing on Syntactic Tree Representation

Microblog, especially Twitter, has become an integral part of our daily life, where millions of users expressing their opinions daily towards various target entities. Detecting and analyzing user stances from such massive opinion-oriented twitter posts provide enormous opportunities to journalists, governments, companies, and other organizations. However, the short length characteristics and frequent use of idiosyncratic abbreviations in tweets make this task challenging to infer the users’ stance automatically towards a particular target. In this paper, we leverage the syntactic tree representation of tweets to detect the tweet stance. We devise a new parts-of-speech (POS) generalization technique and employ the hashtag segmentation for effective tree representation. Then, we make use of support vector machine (SVM) classifier with three different tree kernel functions including subtree (ST) kernel, subset tree (SST) kernel, and partial tree (PT) kernel as the base-classifiers. Finally, a majority voting count based prediction scheme is employed to identify the tweet stance. We conducted our experiments using SemEval-2016 twitter stance detection dataset. Experimental results demonstrate the effectiveness of our proposed method over the baseline and the state-of-the-art related works.