STEM: Unsupervised STructural EMbedding for Stance Detection

TL;DR

This paper introduces an unsupervised, domain-independent structural embedding framework for stance detection that effectively classifies speaker stances in discussions without labeled data.

Contribution

The novel framework constructs interaction networks and derives topological embeddings that distinguish speakers' stances, outperforming or matching supervised models across multiple datasets.

Findings

Embeddings cluster speakers by stance with similar vectors for same stance.

The method achieves comparable or better performance than supervised models.

Structural embeddings relate to speakers' expressed valence.

Abstract

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.