Enhancing the Readability of Automatically Generated Unit Tests with Large Language Models

Abstract

In the pursuit of maintaining high standards in software development, the clarity of unit tests is crucial. This study investigates the application of ChatGPT, a state-of-the-art language model, to enhance the readability of unit tests. An empirical approach was employed, starting with automatically generated JUnit tests, which were then processed by ChatGPT for identifier renaming, variable name renaming, and documentation enhancement. The impact of these AI-driven refinements on readability was assessed using a combination of quantitative readability metrics and qualitative surveys. The evaluation was based on methodologies from DeepTC-Enhancer, Daka's, and TestDescriber. Our results indicate some interest on the part of developers in using ChatGPT to improve the readability of unit tests, making them more understandable.