摘要
由于异常值和缺失值的扰动影响,业务流程系统可能会产生低质量的事件日志.针对缺失活动的日志修复问题,现有日志修复研究主要从缺失活动的重构方面进行展开,很少有从预测缺失活动角度进行修复工作.鉴于此,提出一种结合迹行为特征的卷积神经网络模型,用来修复迹中的缺失活动.其核心思想是按照时间属性和活动属性两个维度,根据活动之间的行为关系,将业务流程的事件日志转换为空间数据,继而转化为图像矩阵,并通过卷积神经网络模型训练来预测缺失活动.所提方法从事件日志的角度出发,不依赖于任何有关生成事件日志的业务流程模型先验知识,最后利用真实和人工生成事件日志将所提方法与已有研究进行对比分析,实验结果表明所提方法在活动修复精确度上优于已有研究结果.
Abstract
Business process systems may produce low-quality event logs due to the disturbance of outliers and missing values.For the problem of missing activity log repairing,existing log repair research mainly focuses on the recon-struction of missing activity,whereas few work is carried out from the perspective of predicting missing activi-ty.Based on Convolutional Neural Networks(CNN)model incorporating the behavioral features of the trace,an ap-proach of repair missing activity in the event logs was investigated.Its core idea was to transform event logs of busi-ness processes into spatial data in terms of both temporal properties and activity properties,and also depending on the behavioral relationship between the activities.The spatial data was further transformed into image matrices and trained by CNN models,which could achieve the aim of predicting missing activity.The purposed method did not depend on any prior knowledge about the business process model except its event logs.The purposed method was compared with the existing research by using two kinds of event logs,named real and artificially generated event logs.Experimental results showed that the purposed method was superior to the existing research results in activity repair accuracy.
基金项目
国家自然科学基金资助项目(61902002)
国家重点研发计划资助项目(2023YFC3807501)
NETL
NSTL
万方数据