For smart devices such as smartphones and tablets, developing new software using open source software (OSS) is becoming mainstream. While OSS-based development can greatly increase project productivity, it is more difficult to identify the cause of software defects. In this paper, we propose a deep learning model that performs unsupervised learning based on the log data accumulated in the project and calculates the degree of abnormality per line for newly given logs. The proposed method is evaluated using open supercomputer system log data, Blue Gene/L, and the accuracy of the proposed method is compared with the conventional log anomaly detection method using LSTM AutoEncoder. As a result of the comparative experiment, it was found that the proposed method performed better than the conventional method in the two scores of AUROC and F1 Score at the cutoff point.
CITATION STYLE
Hirakawa, R., Tominaga, K., & Nakatoh, Y. (2020). Software Log Anomaly Detection Through One Class Clustering of Transformer Encoder Representation. In Communications in Computer and Information Science (Vol. 1224 CCIS, pp. 655–661). Springer. https://doi.org/10.1007/978-3-030-50726-8_85
Mendeley helps you to discover research relevant for your work.