Reproduce “Stance Reasoner: Zero-Shot Stance Detection on Social Media with Explicit Reasoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Maksym Taranukhin, Vered Shwartz, Evangelos Milios
Stance Reasoner prompts a pretrained language model with six labeled examples to connect a post to background knowledge and infer whether it favors, opposes, or is neutral toward an unseen target. Across three Twitter datasets, LLaMA 65B scores 72.6 macro-F1 on SemEval-2016 Task 6a, versus 68.5 for its few-shot chain-of-thought baseline; results do not establish broader transfer or faithful causal reasoning.
Key insights
- Explicit reasoning through Chain-of-Thought (CoT) prompting allows pre-trained language models to perform robust zero-shot stance detection without fine-tuning.
- Pre-trained language models inherently possess sufficient background knowledge for nuanced stance detection when guided by structured reasoning prompts.
- The diversity of reasoning strategies included in the in-context examples is critical for the framework's overall performance and generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.