Language Independent Stance Detection: Social Interaction-based Embeddings and Large Language Models

TL;DR

This work shifts stance detection from text-centric modeling to leveraging social interaction data by introducing Relational Embeddings learned from real user interactions (retweets and friends). When combined with textual classifiers, these embeddings consistently improve performance across seven datasets and four languages, achieving state-of-the-art results in most settings and outperforming several LLM baselines. The method also surpasses popular graph-based approaches like DeepWalk and Node2Vec, underscoring the value of modeling genuine interaction pairs rather than artificial walks. The approach is language-agnostic, facilitates robust cross-dataset generalization, and opens avenues for reducing labeled data needs through user-oriented representations. Datasets and code are made publicly available to support further research in multilingual, interaction-aware NLP tasks.

Abstract

The large majority of the research performed on stance detection has been focused on developing more or less sophisticated text classification systems, even when many benchmarks are based on social network data such as Twitter. This paper aims to take on the stance detection task by placing the emphasis not so much on the text itself but on the interaction data available on social networks. More specifically, we propose a new method to leverage social information such as friends and retweets by generating Relational Embeddings, namely, dense vector representations of interaction pairs. Our experiments on seven publicly available datasets and four different languages (Basque, Catalan, Italian, and Spanish) show that combining our relational embeddings with discriminative textual methods helps to substantially improve performance, obtaining state-of-the-art results for six out of seven evaluation settings, outperforming strong baselines based on Large Language Models, or other popular interaction-based approaches such as DeepWalk or node2vec.

Figures (6)

• Figure 1: One hidden layer artificial neural network.
• Figure 2: Relational Embeddings + SVM model architecture.
• Figure 3: SVM-based combined models architecture.
• Figure 4: Transformer-based combined model architecture.
• Figure 5: Relational embedding representation of training set users (PCA dimension reduction to 2).