Abstract: As the applied changes and adaptability of process model,plenty of process variants are derived from the same model in practical applications.Many similarities in structure and function are existed among these process variants,but at least some details and structures are distinguished from each other.It is difficult to configure and maintain these variants.To use fewer system models to describe process variants,a process variants cluster mining method was proposed.Variant models were transformed into matrices by combining the concept of causal behavioral profiles.Weighted cosine similarity was used to cluster activities and determine the behavior relationship between activities.An objective reference process model was obtained by iterating repeatedly until all activities clustered.The obtained clustered model was easier to configure different process variants than the original reference model.In the simulation experiments,the purposed method was compared with the traditional process mining and classical heuristic algorithm,the experiment results showed the feasibility and effectiveness of the proposed method in mining reference model.

Key words: cluster, cosine similarity, causal behavioral profiles, process mining, process variants, reference model

摘要： 由于应用的更改和流程模型的自适应性,导致现实应用中存在大量来自同一模型的流程变体。这些流程变体在结构功能上存在诸多共同之处,但至少存在一些细节结构相互区分,配置和维护这些流程变体非常困难。为实现使用较少的系统模型来描述流程变体,提出一种流程变体聚类挖掘算法。首先结合因果行为轮廓的概念将变体模型矩阵化,然后应用加权余弦相似度来聚类活动并确定它们之间的行为关系,最后通过反复迭代直到聚类所有活动,得到一个目标参考流程模型,该模型比原参考模型更容易配置不同的流程变体。仿真实验中,将该算法和传统的流程挖掘算法,以及经典的启发式算法进行对比,结果表明了该算法在挖掘参考模型方面的可行性和有效性。

关键词: 聚类, 余弦相似度, 因果行为轮廓, 流程挖掘, 流程变体, 参考模型