.png){kind=icon}
Return
CLlog: A collaborative learning-based anomalous log detection model
.png){kind=icon}
PRE
AI
DOI：10.1016/j.comnet.2025.111943.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Existing supervised learning-based log anomaly detection methods typically require a large number of labeled anomalous samples. However, this reliance on labeled data is costly, while semi-supervised learning models face challenges in effectively extracting features from logs. To address these issues, we propose CLlog, a collaborative learning-based anomalous log detection model. Collaborative learning leverages the mutual learning and knowledge transfer between Graph Attention Networks and Graph Convolutional Networks to achieve better feature extraction and classification. This model does not require labeled anomalous samples, significantly reducing data labeling costs. In our approach, logs are first converted into directed graph structures to capture the complex interactions and network topology features among log entries. Then, through the collaborative learning mechanism, GAT and GCN guide and learn from each other during training, leading to mutual improvements in accuracy. Finally, through the collaborative learning mechanism, our model effectively combines the strengths of GAT and GCN, enabling it to accurately identify anomalous logs while being trained only on normal samples. Experiments show that our model demonstrates significant detection advantages across five public datasets, achieving a 7 % improvement in F1 score. It effectively addresses the challenges of high labeling costs and improving detection accuracy in log anomaly detection.
Journal
IF:
4.6
Papers:
1.8K
Citations:
1.6W
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveShareSaveShareSave
View more