Abstract: Low-frequency behavior mining is one of the important contents of business process management.It is especially important to distinguish effective low-frequency and noise in business process optimization.The existing mining methods mostly study low-frequency behavior from data attributes,and less frequently analyze low-frequency behaviors according to behavior attributes among different modules.On this basis,a method of mining and optimizing the low-frequency behavior of business process was proposed based on Petri nets.The log behavioral relationship was illustrated with the direct flow graph cut by process tree and matched with the initial model to find all the low frequency sequences.The behavior distance between the log and the model was calculated.The effective low-frequency log and noise log was divided based on the behavior closeness to optimize the event log.An optimized business process model was obtained by using the event log without noise sequence to merge the interactive module network and the feature network.The feasibility and effectiveness of the proposed method were verified by concrete example analysis and simulation experiments.

Key words: communication behavior profile, behavioral closeness, low frequency behavior, feature network, fusion, Petri net, business process

摘要： 低频行为的挖掘是业务流程管理的重要内容之一,区分有效低频和噪音在业务流程优化中显得尤为重要。已有挖掘方法多是从数据属性研究低频行为,较少根据不同模块间的行为属性来分析低频行为,由此提出基于Petri网的业务流程低频行为的挖掘与优化方法。首先,通过用流程树切的直接流图表示日志的行为关系,并与初始模型做匹配,发现所有的低频序列;然后,计算日志与模型的行为距离向量,基于行为紧密度区分有效低频日志和噪音日志,优化事件日志;其次,利用不包含噪音序列的事件日志通过融合交互模块网与特征网,挖掘得到一个优化的业务流程模型;最后,通过具体的实例分析和仿真实验验证了该方法的有效性。

关键词: 通讯行为轮廓, 行为紧密度, 低频行为, 特征网, 融合, Petri网, 业务流程