Adapting Large Language Models to Aid End-Users in Bug Reporting

Authors

  • ALYSSA MCGOWAN Department of Computer Science, George Mason University, Fairfax, VA
  • Junayed Mahmud Department of Computer Science, George Mason University, Fairfax, VA and Department of Computer Science, University of Central Florida, Orlando, FL
  • Kevin Moran Department of Computer Science, George Mason University, Fairfax, VA and Department of Computer Science, University of Central Florida, Orlando, FL

DOI:

https://doi.org/10.13021/jssr2023.3932

Abstract

 Bug reporting serves as an integral resource for software maintenance by providing detailed information about issues within an app. However, user-submitted bug reports often lack adequately detailed reproduction steps for developers, prompting the need for a tool that makes use of app information to assist reporters in generating improved bug reports. Furthermore, with tools based on large language models (LLMs) such as ChatGPT on the rise, this project aims to better understand how LLMs can assist end-users in the bug reporting process and develop a tool that leverages LLMs to automatically rewrite low-quality bug reports to contain more detailed and complete information. To help assist LLMs in “comprehending” app interfaces, we developed a system that automatically extracts and formats information such as GUI components and descriptions of app functionality for ChatGPT. Our tool constructs an app execution model using tools for automatically exploring and recording interactions with app UIs and prompts an LLM using chain-of-thought reasoning and few-shot learning to predict reproduction steps. We are currently working to evaluate the efficacy of our tool in improving low-quality bug reports using open-source benchmarks from the research community.

Published

2023-10-27

Issue

Section

College of Engineering and Computing: Department of Computer Science

Categories