This research focuses on stance detection, predicting an author’s viewpoint towards a subject, using Large Language Models (LLMs). Current methods rely on manual annotation and training a supervised machine learning model. However, this process has limitations. This study uses LLMs for stance classification with minimal human labels. The researchers scrutinize four types of prompting schemes combined with LLMs and compare their accuracies with manual stance determination. The results show LLMs can match or sometimes exceed the benchmark results in each dataset, but their overall accuracy is not definitively better than supervised models. This suggests potential areas for improvement in the stance classification for LLMs. The use of LLMs opens up avenues for unsupervised stance detection, reducing the need for manual collection and annotation of stances, and expanding stance detection capabilities across languages.
Publication date: 26 Sep 2023
Project Page: https://anonymous.4open.science/r/LLM-Stance-Labeling/README.md
Paper: https://arxiv.org/pdf/2309.13734
