Identifying Security Bug Reports Based Solely on Report Titles and Noisy Data

Abstract

Identifying security bug reports (SBRs) is a vital step in the software development life-cycle. In supervised machine learning based approaches, it is usual to assume that entire bug reports are available for training and that their labels are noise free. To the best of our knowledge, this is the first study to show that accurate label prediction is possible for SBRs even when solely the title is available and in the presence of label noise.