A Dataset and Service for Automated Identification of Security Bug Reports

Read full text Summarise with VIRAT
Authors: Horácio L. França, César Teixeira, Nuno Laranjeiro
Cited by: 0
References: 50

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Reporting the presence of a security bug allows developers to take action and correct security vulnerabilities before they can be exploited by attackers. The analysis of a security bug report is mostly a manual activity, which, among many other tasks, requires the effort and time of one or more developers to check if the reported issue is actually an issue or not. Several studies have tried to automate report identification using machine learning, but given that vulnerabilities are infrequent bugs, models do not really perform well, especially due to the fact that it is difficult to collect sufficient and diverse data for training. We aim to contribute a large and diverse dataset to aid in security report identification. In this work, we gather security bug reports that are diverse and were collected from heterogeneous sources to form a security bug report dataset. The main goal is to support research that aims at automating the identification and classification of security bug reports. We provide our dataset, as well as a few pre-trained machine learning models, as a publicly available REpresentational State Transfer (REST) service that enables communication between applications over HTTP).We also illustrate the usefulness of our dataset with a set of typical machine learning experiments, which resulted in models achieving an f-score of 67.5% (relatedwork on similar datasets reports average f-scores ranging between 38% and 66.5%), while especially focusing on different aspects, such as understanding which content to use for model training or whichwords are more important in the classification process. The work presented here resulted in a dataset to be used in security bug report identification that is larger, with 206,728 reports, and more heterogeneous, with 9 projects, than those observed in related work. Our case studies show their usefulness and highlight typical concerns regarding the use of models to classify bug reports.

Fields of Study & Subjects

Software Engineering Research Software Testing and Debugging Techniques Web Application Security Vulnerabilities

Publication details

Journal / source
IEEE Access
Volume
14
Pages
46843–46857
Publication date
1 January 2026
Language
EN