›› 2021, Vol. 27 ›› Issue (9): 2661-2669.DOI: 10.13196/j.cims.2021.09.018

Previous Articles     Next Articles

Process model matching based on representation learning

  

  • Online:2021-09-30 Published:2021-09-30
  • Supported by:
    Project supported by the National Natural Science Foundation,China(No.61602278,71704096,U1931207),the Natural Science Foundation of Shandong Province ,China(No.ZR2017MF027,ZR2019LZH001),the Taishan Scholars Program of Shandong Province,China(No.TS20190936),and the Shandong University of Science and Technology Research Fund,China(No.2015TDJH102).

基于表示学习的业务过程模型匹配方法

倪维健,吉桂芳,曾庆田+,刘彤,段华   

  1. 山东科技大学计算机科学与工程学院
  • 基金资助:
    国家自然科学基金资助项目(61602278,71704096,U1931207);山东省自然科学基金资助项目(ZR2017MF027,ZR2019LZH001);山东省泰山学者特聘专家资助项目(TS20190936);山东科技大学优秀科研团队计划资助项目(2015TDJH102)。

Abstract: In recent years,process aware information system has been widely applied in various enterprises and institutions,generating a large number of business process models.The management of massive business process models has become an important task in business process management,in which Process Model Matching (PMM) is one of the key technologies.Most of the PMM approaches only rely on activity tag information and do not yet achieved the desired accuracy.A novel PMM approach based on representation learning was proposed.The activities in a business process model were represented as multi-dimensional real-valued vectors that encoded the context information and relationship information contained in the business process model.A vector space mapping model was proposed to model the activity vectors from different business process models,thus enabling activity matching between process models.Experimental results on public datasets verified that the proposed approach outperformed traditional PMM approaches based on syntax,semantic and structure information.

Key words: representation learning, process model matching, activity vector, business process management

摘要: 近年来过程感知信息系统在各类企事业单位中得到了极大的普及,产生了大量业务过程模型,如何对海量业务过程模型进行有效管理成为业务过程管理领域中的重要任务。业务过程模型匹配是海量业务过程模型管理的一项关键技术,然而大多数过程模型匹配方法仅依赖于活动标签信息,在匹配精度上还有一定欠缺。为此提出一种基于表示学习的过程模型匹配方法。首先基于业务过程模型中蕴含的上下文信息和关系信息将业务过程模型中各个活动表示为多维实数向量;在此基础上,设计了一个向量空间映射模型,将不同过程模型中的活动向量进行映射,进而实现过程模型中的活动匹配。基于公开数据集的实验结果表明,所提方法相对于传统的基于语法、语义、过程模型结构的匹配方法具有一定的优势。

关键词: 表示学习, 过程模型匹配, 活动向量, 业务过程管理

CLC Number: