Krone-viz: Interactive Hierarchical Log Anomaly Analytics with LLM Augmentation
Researchers have introduced Krone and Krone-viz, a novel framework and interactive visualization system designed to enhance log anomaly detection in modern computing systems. Addressing the limitations of unstructured, flat log sequences, Krone employs a hierarchical log abstraction that transforms data into semantically coherent units across entity, action, and status levels. This approach enables precise anomaly detection, localization, and explanation through modular detection tasks and selective Large Language Model (LLM) reasoning. The associated Krone-viz system allows software engineers and system operators to examine hierarchical decompositions, inspect LLM-generated explanations for abnormal segments, and refine knowledge using human-in-the-loop guardrails. Validated on the HDFS benchmark dataset, the tool aims to make complex log analysis more interpretable and actionable. The project, led by Lei Ma and colleagues, provides open-source code and a live demo, representing a significant advancement in applying AI to database management and software engineering diagnostics.
Wire timeline
Krone-viz: Interactive Hierarchical Log Anomaly Analytics with LLM Augmentation
Researchers have introduced Krone and Krone-viz, a novel framework and interactive visualization system designed to enhance log anomaly detection in modern computing systems. Addressing the limitations of unstructured, flat log sequences, Krone employs a hierarchical log abstraction that transforms data into semantically coherent units across entity, action, and status levels. This approach enables precise anomaly detection, localization, and explanation through modular detection tasks and selective Large Language Model (LLM) reasoning. The associated Krone-viz system allows software engineers and system operators to examine hierarchical decompositions, inspect LLM-generated explanations for abnormal segments, and refine knowledge using human-in-the-loop guardrails. Validated on the HDFS benchmark dataset, the tool aims to make complex log analysis more interpretable and actionable. The project, led by Lei Ma and colleagues, provides open-source code and a live demo, representing a significant advancement in applying AI to database management and software engineering diagnostics.
cs.AI updates on arXiv.org