Computer Vision News - September 2022

52 Congrats, Doctor! Log files are semi-structured text files. They are written nowadays by almost any computer system and hold precious information about events that happen during system operation. Following, log file processing is commonly used practice in various domains such as usage analysis, cyber security, error statistics, trend detection, and many more. Log File Changes With modern software development paradigms, software is subject to continuous updates. With code changes also logging statements might be adapted. In fact, 36% of all logging statements undergo change at least once throughout their lifetime. However, common parsing pipelines rely on rigid patterns. As soon as log file structures change, state-of-the-art parsers cannot transform logs successfully anymore. This results in lowqualitydatasets with missing or erroneous values and leads to failing subsequent data analyses. Flexible Parsing Challenges due to ever-changing log structures are overcome with our flexible parser called “FlexParser”. The goal is to extract the desired event values despite event key changes. Therefore, it includes text preprocessing and a deep learning model. Next to standard tokenization and stemming, vectorization is applied using one-hot-encoding to ensure equal distances. The deep learning model poses a stateful Long Short-Term Memory (LSTM). The usefulness of “FlexParser” has been validated in terms of generality (7 publicly available data set), flexibility (7 update scenarios), and competitiveness (16 parsing models). “FlexParser” was found to outperform all state-of-the-art parsers Nadine Rücker has recently completed her PhD with great success at the Pattern Recognition Lab at the University of Erlangen- Nürnberg, Germany. She is driven to build systems that make people’s lives easier. In line with her motivation, her research was about simplifying the process of log file parsing as well as actionable insights generation from logs. Therein, she combined text mining with deep learning methods and set up systems that utilize change to human’s advantage. Next to her PhD, she excelled in her full-time job at Siemens Healthineers. Currently, she continues her work as a software product owner at Siemens Healthineers. Therein, she enables people to make medical fleet management for customers a little bit easier every day.