STEM: Unsupervised STructural EMbedding for Stance Detection
This addresses stance detection for applications like fake news analysis, offering an unsupervised alternative to supervised methods, though it is incremental in leveraging network structures.
The paper tackles stance detection by proposing an unsupervised, domain-independent framework that constructs interaction networks from discussions to derive topological speaker embeddings, which group speakers by stance and represent opposing stances with antipodal vectors. It outperforms or matches supervised models on three datasets and provides confidence levels for outputs.
Stance detection is an important task, supporting many downstream tasks such as discourse parsing and modeling the propagation of fake news, rumors, and science denial. In this paper, we propose a novel framework for stance detection. Our framework is unsupervised and domain-independent. Given a claim and a multi-participant discussion - we construct the interaction network from which we derive topological embedding for each speaker. These speaker embedding enjoy the following property: speakers with the same stance tend to be represented by similar vectors, while antipodal vectors represent speakers with opposing stances. These embedding are then used to divide the speakers into stance-partitions. We evaluate our method on three different datasets from different platforms. Our method outperforms or is comparable with supervised models while providing confidence levels for its output. Furthermore, we demonstrate how the structural embedding relate to the valence expressed by the speakers. Finally, we discuss some limitations inherent to the framework.