Balancing the Imbalance: Enhancing MOOC Discussion Forum Urgency Prediction with LLM-Generated Data Augmentation

Abstract

Discussion forums in Massive Open Online Courses (MOOCs) often contain urgent posts requiring timely instructor intervention, but identifying these posts is challenging due to severe class imbalance in urgency levels. This study presents a novel approach combining Large Language Models (LLMs) and traditional machine learning to address this challenge. We develop a two-stage pipeline that first uses LLMs to generate synthetic forum posts, creating a balanced dataset across seven urgency levels, then evaluates five different machine learning models on both original and balanced datasets. Our results demonstrate significant improvements across all models, with the Support Vector Regression achieving the highest performance (R² improvement from 0.298 to 0.857). Through comprehensive evaluation using multiple metrics and cross-validation, we show that our approach not only improves model performance but also enhances prediction stability across different urgency levels. This work contributes to both educational data mining methodology and practical MOOC forum management.

Materials

• Full BibTeXapi.crossref.org

Cite

@inproceedings{li2025balancing,
  author = {Li, Hongming and Botelho, Anthony F.},
  title = {Balancing the Imbalance: Enhancing MOOC Discussion Forum Urgency Prediction with LLM-Generated Data Augmentation},
  booktitle = {Proceedings of the 18th International Conference on Computer-Supported Collaborative Learning - CSCL 2025},
  year = {2025},
  pages = {455--459},
  publisher = {International Society of the Learning Sciences},
  doi = {10.22318/cscl2025.111336},
  url = {https://doi.org/10.22318/cscl2025.111336}
}