Content Details | NUML Online Research Repository

Content Details

Back to Department Listing

Title DUPLICATE BUG REPORT DETECTION USING DATA AUGMENTATION TECHNIQUE

Author(s) Anisa Tariq

Abstract In software projects, developers, testers, and end users identify bugs and report the bugs to the triager. To manage these bug reports, various Bug Tracking Systems (BTS) such as Bugzilla or Jira are used. One bug may be reported by multiple persons to the system which generate Duplicate Bug Reports in the system. Duplicate Bug Report Detection (DBRD) is very important because it results in a significant depletion of human resources. Many researchers proposed a range of machine learning techniques to detect the duplicate bug reports. The existing techniques performs well when a large number of bug reports are used as a training dataset but the performance of existing techniques significantly decreased for small dataset. To overcome this challenge, the data augmentation technique is used to increase the number of bug reports for the projects having small number of bug reports as training data. Various data augmentation techniques like synonym replacement, random insertion, component shuffling and class balance are used to increase the bug data. To validate the performance of data augmentation technique for duplicate bug detection, we used various deep learning models e.g. CNN, LSTM and BERT. We also compare the results of various deep learning techniques to analyze which model performs better with the augmented bug reports data. Our results show that data augmentation improved the results for all three models in term of accuracy, precision, recall, F1-socore and AUC score. The accuracy achieved on augmented data is 94.77%, 94.77% and 96.30% for LSTM, CNN and BERT respectively.

Type Thesis/Dissertation MS

Faculty Engineering and Computer Science

Department Engineering

Language English

Publication Date

Subject

Publisher

Contributor(s)

Format

Identifier

Source

Relation

Coverage

Rights