Abstract: The execution time of a business process is usually a key assessment indicator.Cases and activities that are not within the specified execution time can be regarded as abnormal process time,which may cause risks such as overtime and customer complaints.Therefore,the root cause analysis of abnormal process time can put forward targeted rectification plans and intervention measures.However,there are many potential causes of time anomaly,which are difficult to extract comprehensively.Moreover,comprehensive analysis seriously affects the execution efficiency and the accuracy of the results.To solve the above problems,to explore the root causes of abnormal business process time,a framework for tracing the root causes of abnormal business process time based on causal inference was proposed.The time information and workload of the event log were expanded to provide rich candidate reasons.The causal hypothesis of abnormal execution time of cases and activities was generated,and the corresponding potential reasons were determined.Then,the causal inference approach based on meta-learning was applied to estimate the causal effect and determine the causal relationship.When the root cause of the abnormal case execution time included an activity execution time,the cause of the abnormal activity execution time was traced back.Finally,it compared with the state of the art approach on five real event logs,and the root cause results were visualized.The experimental results showed that the proposed approach could effectively improve the root cause analysis efficiency of abnormal process time and get more reasonable reasons.

Key words: business process, causal inference, root cause, time anomaly, meta-learning

摘要： 流程的执行时间通常是关键的业务考核指标,未在规定执行时间内完成的案例和活动均可视为流程时间异常,可能导致超时、客户投诉等风险。因此,剖析流程时间异常的根本原因可有针对性地提出整改方案和干预措施。然而,导致时间异常的潜在原因繁多难以全面提取,逐一分析严重影响执行效率,并且根本原因分析的准确性难以保证。针对上述问题,为探究业务流程时间异常的根本原因,提出一种基于因果推断的业务流程时间异常根因溯源分析框架。首先,扩展事件日志的时间信息和工作负载,提供丰富的候选原因。其次,生成案例和活动执行时间异常因果假设,确定对应的潜在原因,建立案例与活动间的联系以便于根因溯源。然后,应用基于元学习的因果推断方法,估计因果效应以确定因果关系。最后,在5个真实事件日志上与最新方法进行比较,并可视化根因结果。实验结果表明,所提方法可有效提高流程时间异常根因的分析效率,得到更合理的原因。

关键词: 业务流程, 因果推断, 根本原因, 时间异常, 元学习