Stance Detection

Stance detection studies how language expresses a position toward a specific target, such as a policy, public figure, or claim, including when that position is implicit or spread across a conversation. Common approaches include fine-tuned transformer classifiers, target-aware models, lexicons, and newer large language model prompting and reasoning methods; self-training helps extend coverage when labeled data is scarce. Benchmarks such as SemEval-2016 and VAST test performance on social media, while cross-target, multilingual, and conversational evaluations probe whether systems transfer to unfamiliar settings. A central challenge is that simple pro, con, and neutral labels can flatten complex attitudes, and extra context or elaborate prompting does not always improve accuracy.