Automatic bug localization using a combination of deep learning and model transformation through node classification

Bug localization is the task of automatically locating suspicious commands in the source code. Many automated bug localization approaches have been proposed for reducing costs and speeding up the bug localization process. These approaches allow developers to focus on critical commands. In this paper...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Software quality journal Ročník 31; číslo 4; s. 1045 - 1063
Hlavní autori: Yousofvand, Leila, Soleimani, Seyfollah, Rafe, Vahid
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: New York Springer US 01.12.2023
Springer Nature B.V
Predmet:
Algorithms
Classification
Compilers
Computer Science
Data Structures and Information Theory
Deep learning
Graph neural networks
Graphs
Interpreters
Labels
Localization
Machine learning
Nodes
Operating Systems
Programming Languages
Software Engineering/Programming and Operating Systems
Source code
Deep learning
Graph neural networks
Bug localization
Node classification
ISSN:0963-9314, 1573-1367
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Bug localization is the task of automatically locating suspicious commands in the source code. Many automated bug localization approaches have been proposed for reducing costs and speeding up the bug localization process. These approaches allow developers to focus on critical commands. In this paper, we propose to treat the bug localization problem as a node classification problem. As in the existing training sets, where whole graphs are labeled as buggy and bug-free, it is required first to label all nodes in each graph. To do this, we use the Gumtree algorithm, which labels the nodes by comparing the buggy graphs with their corresponding fixed graphs. In classification, we propose to use a type of graph neural networks (GNNs), GraphSAGE. The used dataset for training and testing is JavaScript buggy code and their corresponding fixed code. The results demonstrate that the proposed method outperforms other related methods.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0963-9314
1573-1367
DOI:10.1007/s11219-023-09625-5