Action-oriented process mining: bridging the gap between insight

Action-oriented process mining: bridging the gap between insights

and actions

面向行为的过程挖掘：弥合洞察力和行为之间的差距

Abstract

摘要

As business environments become more dynamic and complex, it becomes indispensable

for organizations to objectively analyze business processes, monitor the existing and potential

operational frictions, and take proactive actions to mitigate risks and improve performances.

Process mining provides techniques to extract insightful knowledge of business processes

covering the continuous monitoring of operational processes and the automated execution

of management actions. Based on the framework, we suggest a cube-based action engine

where actions are generated by analyzing monitoring results in a multi-dimensional way. The

framework is implemented as a ProM plug-in and evaluated by conducting experiments on

both artificial and real-life information systems.

随着业务环境变得更加动态和复杂，组织必须客观地分析业务流程，监控现有和潜在的

运营摩擦，并采取主动行为来降低风险和提高绩效。流程挖掘提供了从流程执行期间收集

的事件数据中提取业务流程的深刻知识的技术。此外，还提出了各种方法来支持对过程相

关问题的实时（预测）监控。然而，持续监控的见解与实际流程改进的具体管理行为之间

的联系缺失。面向行为的过程挖掘旨在将从事件数据中提取的知识与行动联系起来。在这

关键词 行动导向流程挖掘·持续运营管理·化事件为行动·持续流程改进

1 Introduction

1 简介

In [1], the term “business process hygiene” was coined. In the same manner that individuals

do regular check-ups to find potential health issues before they become serious problems,

organizations should objectively analyze key processes to identify existing and potential

problems and improve performance and productivity.

在 [1] 中，创造了术语“业务流程卫生”。与个人进行定期检查以在问题变得严重之前发

现潜在健康问题的方式相同，组织应该客观地分析关键流程以识别现有和潜在问题并提高

绩效和生产力。

Despite the efforts to prevent inefficiencies in design-time, many operational frictions still

事实上，已经做出了许多努力来通过重新设计业务流程来确保组织的整体健康。流程

重新设计通常需要对业务流程进行全面和广泛的分析，并且需要对流程进行根本性的更改。

尽管为防止设计时效率低下做出了努力，但在业务流程的执行过程中仍然会出现许多操作

摩擦，从而产生各种异常。例如，使用已知最佳实践标准化的订单到收款流程在实际执行

过程中经常会出现数千种变体，从而在组织中产生各种问题。

In order to deal with the unanticipated operational frictions that may arise during the

execution of business processes, it is imperative to manage business processes in a

continuous manner. To this end, business managers need to continuously identify problems,

monitor the occurrence of the problems, and take proactive actions to deal with possible risks

to the business process. This continuous management of business processes enables to deal

in a responsive and proactive manner.

为了应对业务流程执行过程中可能出现的意外运营摩擦，以持续的方式管理业务流程

势在必行。为此，业务管理者需要不断发现问题，监控问题的发生，并采取主动行为来应

Process mining has provided many useful techniques to support the continuous

management of business processes. First, process discovery, conformance checking, and

process enhancement have enabled the business managers to identify problems by making

the business processes transparent. Moreover, process monitoring techniques have

effectively detected and predicted problems in an online manner [4–6].

hoc, i.e., the “action part” is still missing and outside the scope of today’s process mining tools.

Indeed, for the actual process improvement, it is necessary to turn the insights from process

mining diagnostics to management actions. For instance, when a bottleneck emerges or is

forecasted to arise, business managers should take actions, such as alerting responsible

employees, bypassing the activity, and assigning more resources, alongside the detection and

prediction of it.

然而，解决此类问题的行为选择仍然是非结构化和临时的，即“行为部分”仍然缺失并

且超出了当今流程挖掘工具的范围。事实上，对于实际的流程改进，有必要将洞察力从流

程挖掘诊断转化为管理行为。例如，当瓶颈出现或预计会出现时，业务经理应该采取行为，

例如提醒负责的员工、绕过该活动、分配更多资源，同时检测和预测瓶颈。

event data and evaluates a set of constraints designed using the process knowledge

constraints. Next, the action engine analyzes the constraint instances and produces action

instances describing needed actions to deal with the existing and potential threats to the

business processes. The action instances are automatically triggered in the underlying

information system to make changes in system configurations, or generate alerts through its

messaging systems.

面向行为的流程挖掘旨在通过系统地结合流程挖掘结果和领域知识来解决此类问题，

并自动化管理操作以改进业务流程。图 1 显示了面向行为的过程挖掘的概述。用于诊断的

过程挖掘技术（参见第 2.1 节）用于从事件数据中提取过程知识。约束监视器分析连续的

事件数据流，并评估一组使用过程知识和领域知识设计的约束。结果，它生成描述监控结

果的约束实例。请注意，事件数据中的事件具有时间关系，并且对事件数据的约束通常具

有时间性质。因此，无法使用具有非时间声明性约束的自动数据质量验证技术 [7] 来监控它

们。接下来，行为引擎分析约束实例并生成描述所需行为的行为实例，以处理对业务流程

的现有和潜在威胁。行为实例在底层信息系统中自动触发，以更改系统配置，或通过其消

息系统生成警报。

在本文中，我们提供了以下贡献：

 We have implemented the cube-based action engine as a ProM plug-in.

 我们已经将基于立方体的行为引擎实现为 ProM 插件。

real-life information systems.

 我们已经评估了基于人工和现实生活信息系统的框架的有效性。

The remainder is organized as follows. We present the related work in Sect. 2. Next, we

explain the background covering a motivating example, basic notation, and event data in

Sect. 3. Afterward, we present the general framework for action-oriented process mining and

the constraint cube-based instantiation of the action engine in Sects. 4 and 5. Section 6

presents the implementation of the framework and experiments both in artificial and real-life

information systems. In Sect. 8, we discuss the implications and limitations and conclude the

paper.

其余部分组织如下。我们在第二章中介绍了相关工作。接下来，在第三章中，我们解

释背景，包括第 1 节中的激励示例、基本符号和事件数据。然后，在第四章和第五章中，

我们提出了面向行为的过程挖掘的一般框架和行为引擎的基于约束立方体的实例化。第 6

章中介绍了在人工和现实信息系统中框架和实验的实现。在第 8 章中，我们讨论了影响和

In this section, we first introduce process mining techniques to provide diagnostics used

在本节中，我们首先介绍流程挖掘技术，以提供用于识别改进点的诊断。之后，我们

介绍了运营支持技术，展示了问题检测和预测的见解与改进业务流程的实际行动之间存在

的差距。最后，我们证明需要一种系统的方法来支持持续的流程改进。

2.1 流程挖掘诊断

Action-oriented process mining begins by defining improvement points in business

processes based on process knowledge. Process mining provides techniques to extract

process-centric insights from event data collected by information systems during the

execution of business processes.

The three main categories of process mining diagnostics include process discovery,

conformance checking, and process enhancement. First, process discovery is to automatically

measures (e.g., service time, throughput time, and waiting time) and analyze the bottlenecks

in business processes. For instance, a performance spectrum plots each process step per case

over time, allowing to analyze non-stationarity of performance and synchronization of

different cases over time. Focusing on the organizational perspective of business processes,

social network analysis provides insights into the relationship among such as handover of

works, subcontracting, and working together. In addition, organizational mining analyzes the

roles in organizations that execute a similar set of activities. Furthermore, root cause analysis

enables to find in-depth explanations of risk incidents. In [13], the root-cause of long

throughput times of process instances is analyzed by focusing on the effect of workloads.

第三，流程增强是用附加信息丰富流程模型，以实现性能分析。通过使用时间戳，可

以使用与时间相关的度量（例如，服务时间、吞吐量时间和等待时间）对其进行扩展，并

分析业务流程中的瓶颈。例如，性能谱图绘制了每个案例随时间变化的每个过程步骤，允

许分析性能的非平稳性和不同案例随时间的同步。社交网络分析着眼于业务流程的组织视

角，提供了对工作移交、分包和合作等之间关系的洞察。此外，组织挖掘分析执行一组相

似活动的组织中的角色。此外，根本原因分析能够找到风险事件的深入解释。在 [13] 中，

Instead of providing process-centric insights by analyzing historical data, operational

For monitoring purposes, the conformance checking techniques are lifted to runtime. For

instance, van Zelst et al [4] propose to compute prefix-alignments to detect deviant behaviors.

Burattin et al suggest a generic framework to compute conformance indicators based on

behavioral patterns.

运营支持不是通过分析历史数据来提供以流程为中心的见解，而是旨在通过分析当前

事件数据来影响当前的运营流程以改进流程。在 [3] 中，描述了运营支持的三个核心活动，

即检测、预测和推荐。首先，检测活动通过运行流程实例来分析偏离行为。一致性检查是

这项活动的支持技术。出于监视目的，一致性检查技术被提升到运行时。例如，van Zelst

等人 [4] 提出计算前缀对齐来检测异常行为。 Burattin 等人提出了一个通用框架来计算基于

derive anti-patterns based on a predefined generic set of patterns regarding business

processes. The patterns describe the rules in terms of the occurrence of tasks, their ordering,

and resource assignments. A CEP engine monitors them and detects violations of the rules.

一些方法利用复杂事件处理 (CEP) 和模型的抽象，例如直接或最终继承。 Weidlich 等

人 [15] 提出了一种方法，可以从作为流程模型抽象的行为配置文件中导出事件查询。使用

CEP 引擎监视事件查询。 Awad 等人 [16] 提出了一种基于预定义的关于业务流程的通用模

式集来派生反模式的技术。这些模式根据任务的发生、任务的排序和资源分配来描述规则。

CEP 引擎监视它们并检测违反规则的情况。

In order to verify properties at runtime, Linear Temporal Logic (LTL) is deployed as a

declarative language for describing the properties [17]. For monitoring purposes, more

possible truth-value states are defined such as temporarily satisfied, temporarily violated,

为了在运行时验证属性，线性时间逻辑 (LTL) 被部署为用于描述属性的声明性语言 [17]。

为了监控目的，定义了更多可能的真值状态，例如暂时满足、暂时违反、永久满足或永久

违反。 Maggi 等人 [5] 提出了一种基于 LTL 和彩色自动机的监控技术。全局自动机表示由

局部自动机表示的所有强加约束的合取。它能够识别不同约束之间可能存在的冲突。

In several approaches, rules are represented as graphical notations such as Petri net. In

[6], the constraints are formalized into Petri-net patterns. The patterns are aligned with event

logs to evaluate whether the execution of business processes comply with them. In [18], the

Petri-net patterns for cloud-based business processes are suggested for certifying compliant