Research On Key Technologies Of Case Identification For Unlabelled Event Logs

Process mining aims to extract useful information from event logs to discover,monitor and improve actual business processes.The core of process mining is process discovery,and the quality of process discovery depends on the quality of event logs.The IEEE Working Group has defined five levels of rank for event logs.The highest level event logs are sound,have clear semantics,well defined events,and automatically record events in a systematic and reliable way.Most process mining technologies rely on standardized event logs,that is,each event in the event logs corresponds to a case.However,there is a common problem that the event will not be displayed pointing to the process instance,and case attribute of events in the collected log is missing,that is,the unlabelled event logs.The existing process discovery technologies are all aimed at standard event logs.Therefore,this paper focuses on case identification of unlabelled event logs.The main research contents are as follows:(1)An Unlabelled Event Logs Case Identification Approach.According to the association rule technology in data mining,we mine the dependency between activities from the unlabelled event logs,and the dependency between activities.Then,according to the dependency relationship between activities,we mine the possible activity relationships between activities,namely concurrent relationship,exclusive relationship and loop relationship,and construct dependency graph;Finally,based on the activity relationship and dependency graph,we propose a case identification algorithm for unlabelled event logs and obtain the standard event logs.This approach can effectively identify concurrent structure and loop structure,and improving the accuracy of case identification.(2)An Unlablled Event Logs Case Identification-Based Process Model Approach.This approach takes the process model and the unlabelled event logs as the input.We transform the process model into a process tree,so as to mine the information between activities from the process tree,that is,the relationship between activities,loop activities,start activity set and complete activity set.Then,based on the information between activities,we propose a new case identification algorithm and output the standard event logs.This approach improves the accuracy of case identification technology based on process model and the time performance.(3)Case Application: A Hierarchical Multi-instance Business Processes Model Discovery Approach.Existing process discovery techniques have mined a flat process model that cannot properly handle event logs in multi-instance process scenarios,and sub logs recorded by multi-instance processes are similar to unlabelled event logs.In order to solve this problem,we propose a hierarchical multi-instance business processes model discovery approach.Firstly,this approch mines the nested relationships between activities from the event logs with multi-instance subprocess information and construct a hierarchical event logs.Then,we identify and reconstruct the sub logs with multi-instance information.Finally,we get the hierarchical multi-instance Petri net model.In order to using the existing quality evaluation indicators,we propose a transformation rule to transform the hierarchical multiinstance Petri nets into a flat Petri net.This approach effectively solves the problem that the existing process discovery technology cannot handle the event logs with multi-instance subprocess information,and we have verified the effectiveness and feasibility of the proposed approach through experiments.