Research On Some Key Issues Of Business Process Management Based On Event Log

In recent years,more and more enterprises use specific or universal business process management system to manage business process with the continuous maturity of business process management technology.These business process management systems are closely aligned with the business processes they support and record a large number of event logs that reflect the state of business process execution.These logs contain a wealth of knowledge that can be used to guide and optimize the execution of business processes in future.Even though the event log contains the information regarding the process execution,how to capture and use such information effectively remains a challenge.Firstly,it is difficult to obtain valuable information from the event log of the unstructured business process by using traditional process mining technologies because their instances have complex and changeable features.Secondly,considering the possible collaboration attributes among employees,it is of great significance to measure it and then assign employees(i.e.,staff or performers)to the activities of business process reasonably for improving the execution efficiency of the business process.Finally,it is urgent to get more features that affect the process outcome and then predict the outcome of an on-going process instance(case)effectively and efficiently in predictive process monitoring.To solve the above mentioned challenges,we firstly propose an approach of mining collaboration patterns based on trace alignment and then propose two approaches of staff assignment for different scenarios of business process.One is based on the mined collaboration patterns of entities and the other is for complex business processes with maximum collaboration.Finally,we present an approach for outcome prediction based on deep learning technologies.The main work of this paper is as follows:Firstly,it is difficult to analyze and model the unstructured business process.To solve this problem,we propose an approach of hierarchical collaboration pattern mining HCPM based on trace alignment.The HCPM first converts the historical completed process instances into entity traces by combining its attributes of activity and resource,and then obtains the alignment matrix through the method of entity trace alignment.Afterwards,the matrix is divided into multiple activity blocks to get some sequences of activity blocks.Finally,the collaboration pattern of activities(the first level)and entities(the second level)are obtained by filtering the activity block sequences and entity sequences with the support thresholds of activities and entities respectively.A case study is conducted by applying HCPM to the the event log,generated by pre-setting the probability between activity and staff.The result shows that the collaboration pattern mined by HCPM is consistent with the preset,and verifies its effectiveness.Secondly,it is difficult to measure the cooperative attribute of employee resources according to the event log and then realize staff assignment in business process.To solve the problem,we propose an approach that defines the concept of compatibility to measure the cooperation between two employees when they perform different activities(i.e.different context)and then proposes FOSA method to assign employees based on the mined high collaboration patterns.The FOSA method can quickly match the process to be assigned with the mined patterns that can be used as candidate assignment strategies by defining two kings of codes,and then choose the one with maximum compatibility of the whole process.Experimental results show that the proposed method based on entity compatibility is more effective than that based on employee comopatibility,and the FOSA method performs better in both effectiveness and efficiency compared with other methods.Thirdly,it is difficult to realize the staff assignment for maximum cooperation in complex business processes based on the collaboration attributes of the employee resources.To solve this problem,we propose a method based on the critical path to measure the compatibility of complex business process.Furthermore,we propose two approaches of staff assignment,GHSA and A*SA,which are based on greedy algorithm and A*algorithm respectively.The GHSA method assigns the staff with local optimal compatibility by using heuristic rules,while A*SA method initiates the lower bound of compatibility firstly and lifts the lower bound gradually,and then utilizes it to prune the non-potential partial assignments for finding the global optimal staff assignment.A large number of experiments on real and synthetic datasets demonstrate the characteristics and advantages of the two methods.Finally,the traditional methods for oucome prediction of an on-going case based on an event log suffer the low accuracy and efficiency.To solve this problem,we propose an approach based on deep learning technique of an attention-based bi-directional LSTM neural network Att-Bi-LSTM to construct a classifier.The Att-Bi-LSTM method can capture more contextual features from each historical case in an event log by the bidirectional LSTM network firstly and then assign the corresponding attention weight to these features by using the attention mechanism for getting a global weighted representation of these features.A classifier can be constructed based on this weighted sum of features and the outcome of each case.In addition,we also propose three other approaches to construct classifiers for comparison,which are based on the LSTM neural network(called LSTM),bi-directional LSTM neural network(called Bi-LSTM)and attention-based LSTM neural network(called Att-LSTM).A large number of experiments based on real datasets show that the proposed method Att-Bi-LSTM is much more efficient in real-time prediction and accurate than the existing traditional machine learning methods.