LEKG: A System for Constructing Knowledge Graphs from Log Extraction
Fangrong Wang, Alan Bundy, Xue Li, Ruiqi Zhu, Kwabena Nuamah, Lei Xu, Stefano Mauceri, Jeff Z. Pan · 2021
Logs record system events and status, which help developers and system administrators diagnose run time errors, monitor running status and mine operation patterns [13, 23]. However, logs are complex and weakly linked, making it difficult to diagnose the causes of failures. While recent studies on log knowledge extraction focus on lifting entities from log messages for enriching a background knowledge graph (BKG), they do not involve knowledge reasoning for inferring implicit relations nor guarantee that the knowledge learned from log streams is consistent with the background knowledge. In this preliminary research paper, we present a log extraction approach to log knowledge graph (KG) construction. It includes a novel strategy that utilizes inference rules from a background knowledge graph to learn new triples and validate triples. Also, it implements a local to global strategy to perform reasoning on temporary log instance graphs (LIGs) then on the extended BKG, which significantly reduces query space. Finally we demonstrate the applicability of this approach by a use case in the context of root cause analysis.