A Recommendation Service System Based On Log Data

With the development of data mining and machine learning,we can get much useful information from historical data.For the business process system,it maintains lots of data related to the process executions,especially the event record corresponding to the activity,which can be extracted as event logs.Through analyzing the historical data of process execution,predictive process monitoring technology establish a predictive model to give prediction and recommendation of running case.This paper presents an approach for predicting the outcome of currently running case.The outcome here refers to whether the case can achieve the business goal after execution.By analyzing the control flow data and business data in the event log,the appraoch generates predictive model using clustering-classification algorithm and builds a set of decision trees before the case execution.At runtime,as long as identifying the cluster similar to running case and then the decision tree corresponding to the cluster is queried using the case's business data,the prediction and recommendation are given.Compared with the existing approach,this approach bulids the necessary predictive models in the pre-process phase and directly applies those pre-built models at runtime,which significantly reduces the runtime overhead of the existing methods.At the same time,the appraoch refine the processing of clustering,combine the control flow data with the business flow data to overcome the sensitivity of the existing appraoch for the process model with large structural differences and improve the accuracy of the clustering analysis.First,this paper introduces the related technologies such as process mining,predictive process monitoring,clustering,and decision trees and so on.By comparing the existing approach of predicting the case outcome,this paper presents the predictive framework and gives the concrete realization method of each step including historical traces encoding,representation of business goal,generation of perditive model and how to give prediction and recommendation.Then,this paper designs the overall framework of the recommendation service system,and implements the log preprocessing module,encoding module and model generating and predicting module.Finally,through two real events log to test the function of the system,evaluate the quality of the perdition and time efficiency,verify the advantages of this predictive framework.