STGTP: Enhancing Stance Detection with Graph and Text Embedding through the Post-Graph Model

This paper presents STGTP, which combines graph and text-based techniques for stance detection of social media posts. This model learns from the text in the post and the relationships between the posts due to users conversations and interactions. The proposed model uses the embeddings constructed using the textual content and an interaction graph of the posts. The concatenated embeddings are fed to a CNN module for predicting the stance label. Results over the ConvinceMe and CreateDebate datasets show that our proposed STGTP surpasses existing stance detection models by up to 3% in terms of Macro F1-score.