BushraDBR: An Automatic Approach to Retrieving Duplicate Bug Reports

TL;DR

BushraDBR addresses the pre-entry detection of duplicate bug reports in Bug Tracking Systems by combining Latent Semantic Indexing for textual similarity with Formal Concept Analysis for clustering. It preprocesses BRs, constructs an LSI-based similarity framework, and thresholds similarities at $0.80$ to identify potential duplicates, which FCA then groups into an AOC-poset to guide triage decisions. Validated on multiple Bugzilla datasets, it achieves near-perfect recall and precision in retrieved duplicates, demonstrating a practical path to reduce redundant maintenance work. The approach offers a concrete, text-driven mechanism to curb DBRs before they enter the Bug Repository, with potential for enhancement via machine learning and wider empirical evaluation.

Abstract

A Bug Tracking System (BTS), such as Bugzilla, is generally utilized to track submitted Bug Reports (BRs) for a particular software system. Duplicate Bug Report (DBR) retrieval is the process of obtaining a DBR in the BTS. This process is important to avoid needless work from engineers on DBRs. To prevent wasting engineer resources, such as effort and time, on previously submitted (or duplicate) BRs, it is essential to find and retrieve DBRs as soon as they are submitted by software users. Thus, this paper proposes an automatic approach (called BushraDBR) that aims to assist an engineer (called a triager) to retrieve DBRs and stop the duplicates before they start. Where BushraDBR stands for Bushra Duplicate Bug Reports retrieval process. Therefore, when a new BR is sent to the Bug Repository (BRE), an engineer checks whether it is a duplicate of an existing BR in BRE or not via BushraDBR approach. If it is, the engineer marks it as DBR, and the BR is excluded from consideration for any additional work; otherwise, the BR is added to the BRE. BushraDBR approach relies on Textual Similarity (TS) between the newly submitted BR and the rest of the BRs in BRE to retrieve DBRs. BushraDBR exploits unstructured data from BRs to apply Information Retrieval (IR) methods in an efficient way. BushraDBR approach uses two techniques to retrieve DBRs: Latent Semantic Indexing (LSI) and Formal Concept Analysis (FCA). The originality of BushraDBR is to stop DBRs before they occur by comparing the newly reported BR with the rest of the BRs in the BTS, thus saving time and effort during the Software Maintenance (SM) process. BushraDBR also uniquely retrieves DBR through the use of LSI and FCA techniques. BushraDBR approach had been validated and evaluated on several publicly available data sets from Bugzilla. Experiments show the ability of BushraDBR approach to retrieve DBRs in an efficient manner.

Figures (11)

• Figure 1: An example of a BR from drawing shapes application.
• Figure 2: The workflow for retrieving DBRs via BushraDBR approach.
• Figure 3: The key elements of BurshraDBR approach.
• Figure 4: DBRs describe the same software failure and use similar vocabularies.
• Figure 5: The DBR retrieval process - BushraDBR approach.