基于机器学习的日志函数自动识别方法

发布时间：2018-06-13 03:25

本文选题：日志函数 + 机器学习　；参考：《计算机工程与科学》2017年01期

【摘要】：随着软件规模的不断增长,日志在故障检测中发挥着愈加重要的作用。然而,目前软件日志缺乏统一标准,常受开发人员个人习惯影响,为大规模系统中日志的自动化分析带来了挑战。其中,日志函数的识别作为日志分析的前提条件,对分析结果有着直接影响。提出了一种基于机器学习的方法以支持日志自动识别。通过系统分析广泛使用的大规模开源软件,总结出日志函数编写的主要形式,并提取不同形式间的共性特征,进而基于机器学习实现了自动日志识别工具iLog。实验显示,使用iLog识别的日志函数能力平均为使用特定关键字的76倍,十折交叉验证得到iLog的分析结果的F-Score为0.93。
[Abstract]:With the increasing of software scale, log plays an increasingly important role in fault detection. However, the lack of a unified standard for software logging, often influenced by the developer's personal habits, poses a challenge to the automated analysis of logs in large-scale systems. The recognition of log function as a precondition of log analysis has a direct impact on the analysis results. A method based on machine learning is proposed to support automatic log recognition. Through the systematic analysis of the widely used open source software, the main forms of writing log functions are summarized, and the common features among different forms are extracted, and then the automatic log recognition tool iLog is implemented based on machine learning. The experimental results show that the ability of using iLog to recognize log functions is 76 times that of using specific keywords on average, and the F-Score of 10% cross-validation of iLog analysis results is 0.93.
【作者单位】：国防科学技术大学计算机学院;
【基金】：国家自然科学基金(61379146,61272483) 腾讯高校合作项目“面向故障检测的大规模开源软件日志增强技术研究”
【分类号】：TP311.53;TP181

【参考文献】