Fine-tuning GPT-3 for automatic scoring in learning progression

• In Event: Learning Progressions

Abstract

Generative Pre-trained Transformer (GPT) is powerful in natural language processing. This research investigates its potential in automatic scoring for items from a learning progression. We leverage the text classification ability of GPT-3 by fine-tuning the models. Results show the fine-tuned GPT-3 models perform satisfactorily in automatic scoring.

Authors

• Mingfeng Xue, University of California - Berkeley

• Mark Wilson, Berkeley School of Education, UC BERKELEY

• Linda Morell, UNIVERSITY OF CALIFORNIA - BER