关键词: 流程挖掘, 模型质量, 噪声日志过滤方法, 依赖关联度, 业务流程管理

Abstract: The behavior with low frequency in the log corresponds to some unnecessary structures in the process model,which leads to the decline of the fitness and accuracy of mining models.To address this problem,a noise log filtering method based on dependent relevancy named Dependency Relevancy driven noise Filtering(DRF)was proposed.According to the statistical frequency of events and edges,the local correlation degree and global correlation degree of the dependency relationship were defined and normalized into a mixed relationship degree,based on which the noise log was filtered out.The noise in the log by the path accessibility analysis was removed,so the other behaviors recorded in the log trajectory were preserved to the maximum extent.Unlike the traditional log filtering algorithms that filtered out the whole log trajectory containing noisy log,DRF could preserve non-noisy log while removing the noisy one.

Key words: process mining, model quality, noisy log filtering method, dependency relevancy, business process management