Replication Package for “Robust and Efficient Log Anomaly Detection: A Hybrid ID-Semantic Approach for Evolving Systems”
Description
This dataset contains the replication package for the article “Robust and Efficient Log Anomaly Detection: A Hybrid ID-Semantic Approach for Evolving Systems.” It includes the extended Deeploglizer-based implementation used in the experiments, the noise-generation scripts used to create template-evolution test data, the LLM prompt used for generating evolved templates, and the processed data/files required to reproduce the reported results on the HDFS and BGL datasets. The package supports three experimental settings within a unified LSTM-based pipeline: (1) the original ID-only baseline, (2) a semantic-only SBERT baseline, and (3) the proposed hybrid ID-semantic approach with OOV semantic fallback. README files are provided to explain the folder structure, installation steps, data preparation workflow, and experiment commands.
Files
Steps to reproduce
Please refer to the README files in extended_deeploglizer/ and noise_generation/ for full reproduction instructions, including installation, noise generation, and experiment commands.