StanceEval 2026: The Second Stance Detection Shared Task

2026-10-02

AI Summary

This paper addresses the challenge of cross-target generalization in Arabic social media stance detection by organizing the second shared task, which systematically evaluates model generalizability across topic-related and unseen targets. The task comprises two tracks designed to assess model robustness under complex linguistic conditions, including extreme polarization, class imbalance, and dialectal sarcasm. Participating systems integrate pretrained model fine-tuning, large language model prompt engineering, retrieval-augmented generation, and hybrid cascaded architectures. Results demonstrate that the top-performing system achieves an F1 score of 0.94, substantially surpassing baselines. Notably, the findings reveal a counterintuitive phenomenon wherein performance on unseen targets exceeds that on related targets, offering critical benchmarks and novel insights for advancing cross-domain stance detection research.

Abstract

StanceEval 2026 is the second edition of the StanceEval shared task series on stance detection in Arabic social media text. Stance detection aims to identify a writer's stance toward a given topic. Given a tweet and a target, participating systems must determine whether the writer's stance is Favor, Against, or None. This edition focuses on cross-target generalization across two distinct evaluation tracks: Track 1 evaluates thematically related cross-target transfer (testing on Women Driving, related to Women Empowerment from training data), while Track 2 evaluates cross-domain transfer to completely unseen targets (E-Cars and Trimester System). The shared task attracted 80 registered teams from 12 countries. During the evaluation phase, 30 unique teams submitted entries, with 21 teams officially ranked in Track 1 and 13 in Track 2 following validation filtering, and 20 teams submitting system-description papers. Participating teams employed diverse methodologies, including fine-tuned pretrained language models, prompt-based and retrieval-augmented large language models (LLMs), fine-tuned LLMs, and hybrid cascades. Top systems achieved impressive $F_{avg2}$ scores of 0.8994 on Track 1 and 0.9400 on Track 2, substantially outperforming the strongest baselines (0.7366 and 0.7475, respectively), where $F_{avg2}$ denotes the macro-averaged F1 score over the Favor and Against classes. Counterintuitively, performance on the unseen targets was higher than on the related target, a disparity could be driven by extreme target polarization, class imbalance, and dialectal or sarcastic nuance across topics.

Problem

• Stance Detection
• Arabic Social Media
• Cross-target Generalization
• Cross-domain Transfer

Innovation

• Stance Detection
• Cross-target Generalization
• Large Language Models
• Retrieval-Augmented Generation
• Arabic Social Media