Detecting Duplicate Workflow Tasks And Noise Logs To Support Process Modeling

Nowadays most information systems employ the defined workflow models to describe the relations between tasks and maintain the entire business processes.With the help of process mining technology,we can find out the actual working models from log data created by information systems,reflecting the actual running states and helping developers to improve it.In actual process modeling,process managers may give different tasks in a process with the same name to reflect its execution in different steps for a special activity.These tasks with the same name are called duplicated tasks.Furthermore,the abnormal running of information systems will create error event logs called noise.Because of the influence of duplicate tasks and noises,it becomes more difficult to obtain a precise process model by simple process mining algorithms.In order to effectively solve the negative effects,we propose two algorithms to solve duplicate tasks and noises before the process mining begins.In order to detect the duplicate tasks in logs,we propose an approach based on casual matrix to overcome the defect that existing algorithms based on event processor and similarity can't detect duplicate tasks appropriately.At first the algorithm analyzes the characteristics of successors in loop and parallel structures and remarks the events.Afterwards it determines the categories and quantities of duplicated tasks by casual matrix.It finally clusters events in the order of similarity,and selects the best result as the final logs.In addition,to detect the noises in logs,we propose an approach based on dependent relevancy.The algorithm concerns both the local relevancy and global relevancy and obtains mixed dependent relevancy by mixing them up.Traditional noise filtering algorithms take single trace as noise unit,leading to losing lots of useful information.In order to filter out noises in a fine-grain level,we propose the noise filtering approach based on reachability analysis,which can leave behind more useful information in logs while filtering out fine-grain noises.We exploit our approach in process mining platform called Pro M.The results demonstrate that our approach detects duplicate tasks more effectively than the congeneric algorithms with respect of precision and simplicity.Meanwhile,our approach in noise filtering can improve the mined process models with higher fitness and precision.