Institutional Repository of Vilnius University
Duplicate bug report detection using siamese bert networks /
Abstract
dc:descriptionBug reporting, tracking and fixing is a major part of every software’s life-cycle. With the growing size, complexity and user base of the software project duplicate bug reports inevitably appear in the defect tracking systems. This introduces a great need for an automated solution for duplicate bug report detection to reduce the cost of software maintenance. Because of that, the natural language processing task of duplicate bug report detection is a popular software engineering challenge quite well studied in the literature. The most recent work focuses on applying separate word embedding training and utilizing deep learning techniques by either using recurrent neural networks with LSTM layers or convolutional neural networks. However, recent breakthroughs in natural language processing with an introduction of Transformer networks and specifically BERT Deep Bidirectional Transformers suggest novel approaches for duplicate bug report detection tasks. In this research, a novel DBRD DistilRoBERTa duplicate bug report detection model is proposed based on Sentence-BERT (BERT adaptations for siamese networks) and DistilRoBERTa (robust optimizations of the BERT model). Novel Unity Technologies bug dataset is introduced for model evaluation accompanied by widely used open bug datasets from Open Office and Eclipse. The model can match the performance of other high-performing complex deep learning models while also achieving great computational efficiency which is a strong advantage and a novel evaluation criterion for duplicate bug reports detection models not highlighted in previous research.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kralikas, Kasparas,
- Contributors dc:contributor
-
- Valaitis, Vytautas
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD107118689&prefLang=en_US