Fine-Tuning Generative Models for Text Summarization with Limited Data - A Practical Guide

• In Event: AERA Roundtable Session Thursday 4:15 pm - Gold 4
  In Roundtable Session: RT6: Tutorials, Guides and Workflows in Quantitative Education Research: Putting our Quantitative Advances into Practical Use

Abstract

The rapid increase in scientific publications across disciplines has made it harder for researchers to stay up-to-date with the latest literature. They often spend hours skimming through papers just to find key information or determine whether a study is relevant to their needs. This study introduces a practical method for fine-tuning pretrained generative language models to automatically generate structured summaries from scientific articles. This helps researchers not only find relevant content more quickly but also understand study details more effectively. Using the PICO framework and including elements - Sample Size, Design Description, and Statistical Analysis as a case example, we examine how such a model can be fine-tuned on limited, manually annotated data using lightweight methods (hyperparameter tuning and generative adapters).

Authors

• Zhifei Li, University of Iowa

• Li Zhu, University of Iowa

• Ariel M. Aloe, University of Iowa

• Kishlay Jha, University of Iowa