Exploring the Feasibility of Retrieval-Augmented Generation for Software Defect Analysis: An Empirical Study

Abstract

A defect database is a structured collection of information about software bugs, errors, and their resolutions throughout the development lifecycle. These databases have valuable information for improving software quality, but their growing volume and complexity make manual analysis inefficient and error-prone. This paper presents an empirical study using real-world data investigating the feasibility of Retrieval-Augmented Generation (RAG) to support interactive exploration of software defect repositories. The study is structured with reference to selected phases of the CRISP-DM framework, which was used as an organizational guideline. Using a real-world software defect database as the empirical basis, the approach automates insight extraction from defect repositories, aiming to improve decision-making in quality assurance and software development by enabling scalable and intelligent exploration of software failures.