基于分层Transformer的事件日志修复  

Event Log Repair Based on Hierarchical Transformer

作　　者：徐猛 卢可 XU Meng;LU Ke(School of Mathematics and Big Data,Anhui University of Science and Technology,Huainan 232001,China;Anhui Engineering Laboratory for Big Data Analysis and Early Warning Technology of Coal Mine Safety,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学数学与大数据学院,安徽淮南232001 [2]安徽理工大学安徽省煤矿安全大数据分析与预警技术工程实验室,安徽淮南232001

出　　处：《榆林学院学报》2025年第2期93-100,共8页Journal of Yulin University

基　　金：国家自然科学基金资助项目(61572035,61402011);安徽省重点研究与开发计划项目(2022a05020005)。

摘　　要：数据缺失是导致事件日志质量低下的主要原因,这种缺失事件日志会造成一定程度的流程偏差,严重制约了后续流程分析的深度与精度。针对事件日志中缺失活动的修复问题,现有研究大多从单一维度特征展开修复工作,在多层次特征挖掘和多维度信息利用方面尚显不足。因此,本文提出了一种基于分层Transformer模型的缺失活动事件日志修复方法。该方法的核心在于构建分层特征提取模块,从事件属性关联起始,扩展至局部行为依赖,最终整合完整上下文执行语义,以实现融合多维度特征信息修复日志缺失活动。通过5个公开可用数据集实验评估,结果显示所提出的方法可以有效修复缺失活动事件日志。In the field of process mining,the completeness of event logs in real business process environments is often difficult to guarantee,leading to deviations between the process models mined from such incomplete event logs and the actual business processes.This significantly hinders the depth and accuracy of subsequent process analysis.To address this issue,existing research primarily have focused on predicting missing data using event attribute information or contextual information within event logs.However,these methods are limited by their simplistic understanding of the underlying data structure and struggle to leverage multi-level,multi-dimensional information.Therefore,this paper proposes an event log repair method based on a hierarchical Transformer model.This method constructs a hierarchical and progressive information aggregation mechanism,starting from the data correlations of event attributes,expanding to the local behavioral dependencies of subsequence events,and ultimately integrating the complete contextual semantics of instances to achieve precise prediction of missing activity labels.Experiments conducted on five public datasets demonstrate that this method performs well in repairing missing activity labels in event logs.

关 键 词：事件日志 缺失活动 分层Transformer 多维度特征融合 日志修复 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]