1488 - 利用支持性洞察的 AI 驱动型持续产品质量改进系统.pdf

1、IBM TechXchange 2025 conference 1488 AI driven continuous Product quality improvement system using support insightsAshish Ghodasara,IBMGias Uddin,York UniversityIBM TechXchange|2025 IBM Corporation1#IBMTechXchangeTable of ContentsSynLog+:Our Optimized Log Parser Empirical study of the effectiveness

2、log parsing techniques for observabilityGeLL:Our Generalizable Log ParserOngoing Work:Design and Deployment of a Log based Event Observability Dashboard 3172327IBM TechXchange|2025 IBM Corporation2Log files are generated by software systemsProvide valuable insight into runtime behavior and anomalies

3、Often unstructured and too large for manual analysisMany logs originate from the same log templateEmpirical study of the effectiveness of log parsing techniques for observability supportLog FilesIBM TechXchange|2025 IBM Corporation3Same templateLog ParsingIBM TechXchange|2025 IBM Corporation4Log par

4、sing is the process of transforming unstructured log messages into structured log dataThe log messages are first parsed to extract the header information,e.g.,date and time,process and thread ID,etc.The rest of the log message is then parsed to identify the constant and variable partsThe constants f

5、orm the log template or event templateThe variables are marked by and constitute the parametersLog ParsingIBM TechXchange|2025 IBM Corporation5Log ParsersIBM TechXchange|2025 IBM Corporation6Based on the design architecture,log parsers can be categorized into:Single-phaseDouble-phaseBased on the tec

6、hniques used to analyze the tokens,log parsers are of two types:Syntax-basedSemantic-basedSyntax-based parsers apply statistical and heuristic techniques to group similar logs together and identify the template of each log groupSemantic-based parsers utilize DL models or LLMs to classify the tokens