Argument Mining in Education: Exploring the Potential of Open-source Small LLMs for Argument Classification and Assessment

Argument mining algorithms analyze the argumentative structure of essays, making them a valuable tool for enhancing education by providing targeted feedback about the students’ argumentation skills. While current methods often use Encoder or Encoder-Decoder deep learning architectures, Decoder-only models remain largely unexplored, offering a promising research direction. In this paper, we propose leveraging open-source, small Large Language Models (LLMs) –such as Llama 3.1 8B– for argument mining through few-shot prompting and fine-tuning, to classify argument types and assess their quality in student essays. Their small size and open-source nature ensure greater accessibility, privacy, and computational efficiency, enabling schools and educators to adopt and deploy them locally. We empirically evaluate the proposed method using the “Feedback Prize – Predicting Effective Arguments” dataset, which contains essays from students in grades 6-12. We demonstrate that fine-tuned small LLMs outperform baseline methods in determining the argument types while achieving comparable performance to the baselines in assessing quality. This work illustrates the educational potential of small LLMs to deliver real-time, personalized feedback, fostering independent learning, and improved writing skills in students while maintaining low computational demands and prioritizing privacy.