Exploring Tutoring Strategies and Dropout Patterns in Online Math Learning via Big Data and LLMs

• In Event: Supporting Student Learning in Online Settings

Abstract

Understanding factors that influence persistence in online math learning is critical for improving student success. This study examined how tutoring strategies in online discussion forums affect dropout in K–12 Algebra I. We first manually coded 24,000 posts and applied data augmentation using Meta LLaMA 3.1 to address class imbalance. After benchmarking multiple models, the BERT classifier achieved the best performance and was used to classify five million discussion posts. Survival analysis was then conducted to explore the effects of tutoring strategies, participation metrics, and linguistic features on dropout risk. Results show that cognitive scaffolding and metacognitive support decreased dropout significantly, but affective support had no quantifiable impact. These insights can guide instructional design and teaching practices in online education.

Authors

• Gokhan Gulfidan, University of Utah

• Chenglu Li, University of Utah

• Bailing Lyu, University of Utah

• Linlin Wu, University of Utah