VIABLE Lab Presents at ISLS 2025 in Helsinki
Chip Li and Dr. Anthony Botelho presented research on LLM-generated data augmentation for MOOC forum urgency prediction at the ISLS Annual Meeting 2025, participated virtually from the lab.
We are excited to share that Chip Li represented the VIABLE Lab at the International Society of the Learning Sciences Annual Meeting (ISLS 2025) held in Helsinki, Finland from June 10-13, 2025. Due to travel constraints, our team participated virtually in this prestigious international conference.
Conference Participation
ISLS 2025, hosted by the University of Helsinki, brought together researchers from around the world under the theme "Educating for world-making: Envisioning and enacting sustainable solutions to global crises." The conference featured both the International Conference of the Learning Sciences (ICLS) and the International Conference on Computer-Supported Collaborative Learning (CSCL) programs.
Research Presentation
Our team presented important work on addressing data imbalance challenges in educational data mining, specifically focusing on MOOC discussion forum urgency prediction. The research demonstrates how Large Language Models can be effectively used to augment educational datasets while maintaining semantic coherence.
Published Research
Paper: "Balancing the Imbalance: Enhancing MOOC Discussion Forum Urgency Prediction with LLM-Generated Data Augmentation"
Authors: Li, H.*, & Botelho, A. F.* (2025, June)
Published in: Computer-Supported Collaborative Learning (CSCL) at the International Society of the Learning Sciences Annual Meeting (ISLS 2025), Helsinki, Finland
Research Impact
This work addresses a critical challenge in MOOC education: identifying discussion forum posts that require urgent instructor intervention. The research demonstrates significant improvements in prediction accuracy, with Support Vector Regression achieving an R² improvement from 0.298 to 0.857 through our LLM-based data augmentation approach.
Key Findings:
- LLM-generated synthetic data effectively addresses class imbalance in educational datasets
- Dramatic improvements in detecting high-urgency posts requiring immediate attention
- Up to 80% RMSE reduction for critical urgency level predictions
- Enhanced model stability and generalization across different urgency categories
Virtual Participation Experience
While we would have loved to experience Helsinki's vibrant academic atmosphere and Finland's renowned hospitality in person, the virtual format allowed us to engage with the global learning sciences community and share our research with international colleagues. The conference's focus on sustainable solutions to global challenges resonated strongly with our lab's mission of developing responsible AI technologies for education.
We look forward to future opportunities to participate in ISLS conferences and continue contributing to the advancement of learning sciences research!