Fine-tuning BERT-based classifiers on LLM-augmented student-response data outperforms unaugmented training for all tested concepts¶

Claim · All claims
Evidence · 1 study · causal r2 · q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Models trained on LLM-augmented data outperformed unaugmented models for all seven concepts across four BERT-based classifiers. → Keith Cochran 2025

Evidence¶

Keith Cochran 2025¶

Keith Cochran, Clayton Cohn, Jean-François Rouet, Peter Hastings. (2025). Comparing Text Augmentation by GPT-3.5 and Llama3 for Evaluating Student Responses. International Journal of Artificial Intelligence in Education. https://doi.org/10.1007/s40593-025-00473-x

q2 · i? · causal · r2

Systematic fine-tuning experiment (15,120 BERT-based models over two datasets, six augmentation methods, ten seeds), comparing LLM-augmented training against an unaugmented BERT baseline using micro-F1. The study "outperformed the unaugmented models for all seven concepts".

"The current study showed that models trained on LLM-augmented data outperformed the unaugmented models for all seven concepts, no matter which of the four classifiers was used."

Discussion¶

Related Claims¶

• LLM-generated responses outperform self-augmentation when augmenting small educational classification datasets — related
• EduBehaviors underperforms the fine-tuned RoBERTa encoder trained on expert-labeled TalkMoves data — related
• Augmentation levels below 20x produce unreliable classification performance in small educational datasets — related
• Datasets with entropy below 0.5 are difficult to augment enough to beat majority-label guessing — related