Reproduce “Enhancing Student Performance Prediction on Learnersourced Questions with SGNN-LLM Synergy” 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
Lin Ni, Sijie Wang, Zeyu Zhang, Xiaoxuan Li, Xianda Zheng, Paul Denny, Jiamou Liu
An international research team developed the Large Language Model enhanced Signed Bipartite graph Contrastive Learning (LLM-SBCL) model, which accurately predicts student performance on learnersourced questions by robustly handling noisy interaction data and addressing the cold start problem for new questions.
Key insights
- Contrastive learning applied to signed bipartite graphs effectively enhances the robustness of student and question representations against noise in learnersourced data.
- Large Language Models can extract valuable intrinsic semantic knowledge from question content, significantly mitigating the cold start problem for new questions.
- The synergy of structural (GNN) and semantic (LLM) information provides a more comprehensive and accurate model for student performance prediction than either approach alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.