Junwei Zhou

Junwei Zhou

Prof., WHUT

Posts by Collection

portfolio

publications

DeepSyslog: Deep Anomaly Detection on Syslog Using Sentence Embedding and Metadata

Published in IEEE Transactions on Information Forensics and Security, 2022

In this paper, we propose DeepSyslog, a deep anomaly detection method for Syslog that integrates unsupervised sentence embedding with event metadata to capture contextual semantics and improve detection accuracy, outperforming existing log-based approaches.

Recommended citation: J. Zhou, Y. Qian, Q. Zou, P. Liu and J. Xiang, "DeepSyslog: Deep Anomaly Detection on Syslog Using Sentence Embedding and Metadata," in IEEE Transactions on Information Forensics and Security, vol. 17, pp. 3051-3061, 2022, doi: 10.1109/TIFS.2022.3201379. keywords: {Metadata;Anomaly detection;Feature extraction;Semantics;Indexes;History;Event detection;Anomaly detection;sentence embedding;event metadata}, https://ieeexplore.ieee.org/document/9865986

LogDLR: Unsupervised Cross-System Log Anomaly Detection Through Domain-Invariant Latent Representation

Published in IEEE Transactions on Dependable and Secure Computing , 2025

In this paper, we propose LogDLR, a novel unsupervised cross-system log anomaly detection method. It uses universal sentence embeddings and a Transformer-based autoencoder to extract domain-invariant latent representations, adapts to heterogeneous log formats, captures semantic dependencies, and achieves efficient and accurate anomaly detection across different systems.

Recommended citation: J. Zhou et al., "LogDLR: Unsupervised Cross-System Log Anomaly Detection Through Domain-Invariant Latent Representation," in IEEE Transactions on Dependable and Secure Computing, vol. 22, no. 4, pp. 4456-4471, July-Aug. 2025, doi: 10.1109/TDSC.2025.3548050. keywords: {Semantics;Anomaly detection;Feature extraction;Transformers;Data models;Training;Autoencoders;Syntactics;Vectors;Long short term memory;Anomaly detection;log analysis;adversarial training}, [https://ieeexplore.ieee.org/document/9865986](https://ieeexplore.ieee.org/document/10910216)

Poster: GLog: Self-Evolving Log Anomaly Type Prediction via Instruction-Tuned LLM and Clustering

Published in Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security, 2025

In this paper, we propose GLog, an end-to-end self-evolving log anomaly prediction framework. It fine-tunes instruction-tuned LLMs to achieve high-accuracy anomaly detection on raw unparsed logs, clusters anomalies for automatic pseudo label generation, and supports continuous self-evolving model optimization, which greatly reduces manual annotation cost and adapts to evolving system behaviors.

Recommended citation: JunWei Zhou, Yuyang Gao, Cheng Tan, Yanchao Yang, and Jianwen Xiang. 2025. Poster: GLog: Self-Evolving Log Anomaly Type Prediction via Instruction-Tuned LLM and Clustering. In Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security (CCS 25). Association for Computing Machinery, New York, NY, USA, 4791–4793. https://dl.acm.org/doi/10.1145/3719027.3760727

talks

teaching

Teaching experience 1

Undergraduate course, University 1, Department, 2014

This is a description of a teaching experience. You can use markdown like any other post.

Teaching experience 2

Workshop, University 1, Department, 2015

This is a description of a teaching experience. You can use markdown like any other post.