计算机应用与软件2026，Vol.43Issue(3)：354-360,368,8.DOI:10.3969/j.issn.1000-386x.2026.03.048
基于半监督主动学习的日志异常检测方法
A LOG ANOMALY DETECTION METHOD BASED ON SEMI-SUPERVISED ACTIVE LEARNING
吴茜雅 1张晨曦 1彭鑫1
作者信息
- 1. 复旦大学计算机科学技术学院 上海 200433||上海市数据科学重点实验室(复旦大学) 上海 200433
- 折叠
摘要
Abstract
In order to ensure the reliability of complex systems,log-based anomaly detection methods have become the focus of research.Existing supervised methods for log anomaly detection require a large amount of labeled data for training,and semi-supervised methods are easily negatively affected by noisy data,and cannot effectively deal with the performance degradation caused by log concept shift.In response to this situation,a log anomaly detection method based on semi-supervised active learning(SSLALog)is proposed,using a supervised Transformer anomaly classification model,and training the model through a combination of semi-supervised self-training learning and active learning.Experimental results show that it outperforms other semi-supervised methods on F1 score,and the practicality is further confirmed by discussing the efficiency.关键词
日志/日志异常检测/半监督学习/主动学习/伪标签Key words
Log/Log anomaly detection/Semi-supervised learning/Active learning/Pseudo-labelling分类
信息技术与安全科学引用本文复制引用
吴茜雅,张晨曦,彭鑫..基于半监督主动学习的日志异常检测方法[J].计算机应用与软件,2026,43(3):354-360,368,8.
