TY - GEN
T1 - Extracting features related to bug fixing time of bug reports by deep learning and gradient-based visualization
AU - Noyori, Yuki
AU - Washizaki, Hironori
AU - Fukazawa, Yoshiaki
AU - Ooshima, Keishi
AU - Kanuka, Hideyuki
AU - Nojiri, Shuhei
AU - Tsuchiya, Ryosuke
N1 - Funding Information:
We thank Prof. Foutse Khomh, Prof. Yann-Gael Gueheneuc and Mr. Qicong Liu for their assistance and useful discussions.
Publisher Copyright:
© 2021 IEEE.
PY - 2021/6/28
Y1 - 2021/6/28
N2 - A bug report is a document indicating when a bug occurs. Developers discuss and resolve the bug through comments in the report. The time required to fix a bug can depend on the bug report. Although many studies have researched bug reports, few have examined bug report comments. Herein we adopt a convolutional neural network (CNN), which is a class of deep neural networks, to classify bug reports into those with short and long fixing times based on the data collected from a bug tracking system. Then we extract the features related to the bug fixing time by visualizing the decision basis that the CNN model uses in the prediction process. We employ a gradient-based visualization technique called Grad-cam to visualize the word sequence that the CNN model uses in the prediction. We use the top ten word sequences as the decision basis to extract the features of the bug report. An experiment confirmed that our method classified more than 36, 000 actual bug reports taken from Bugzilla by short and long fixing times with 75-80% accuracy. Further visualization using Grad-cam shows the difference in the stack trace and the degree of abstraction of the words used. Bug reports with a short bug fixing time are specific and informative with regard to stack trace descriptions. In contrast, those with a long bug fixing time are abstract.
AB - A bug report is a document indicating when a bug occurs. Developers discuss and resolve the bug through comments in the report. The time required to fix a bug can depend on the bug report. Although many studies have researched bug reports, few have examined bug report comments. Herein we adopt a convolutional neural network (CNN), which is a class of deep neural networks, to classify bug reports into those with short and long fixing times based on the data collected from a bug tracking system. Then we extract the features related to the bug fixing time by visualizing the decision basis that the CNN model uses in the prediction process. We employ a gradient-based visualization technique called Grad-cam to visualize the word sequence that the CNN model uses in the prediction. We use the top ten word sequences as the decision basis to extract the features of the bug report. An experiment confirmed that our method classified more than 36, 000 actual bug reports taken from Bugzilla by short and long fixing times with 75-80% accuracy. Further visualization using Grad-cam shows the difference in the stack trace and the degree of abstraction of the words used. Bug reports with a short bug fixing time are specific and informative with regard to stack trace descriptions. In contrast, those with a long bug fixing time are abstract.
KW - Deep learning
KW - Grad-cam
KW - OSS
KW - bug report
UR - http://www.scopus.com/inward/record.url?scp=85114557904&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85114557904&partnerID=8YFLogxK
U2 - 10.1109/ICAICA52286.2021.9498236
DO - 10.1109/ICAICA52286.2021.9498236
M3 - Conference contribution
AN - SCOPUS:85114557904
T3 - 2021 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2021
SP - 402
EP - 407
BT - 2021 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2021
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2021 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2021
Y2 - 28 June 2021 through 30 June 2021
ER -