Research On The Methods Of Business Process Comparison And Merging Based On BPG

With the rapid growth of business process scale and complexity,the analysis and processing of business processes automatically are becoming more and more important.The comparison and merging are the two key aspects of business process analysis and processing.The comparison of the business models can quantitatively measure the similarity between the models,so as to realize the model retrieval.On the other hand,we can identify the same and different parts of the models,and then realize the reconstruction and optimization of the models.Model merging is the integration of multiple business models,which is of practical significance in enterprise merger and reorganization.And the merging of business models is also an application integration method based on business layer.In this paper,we research the method of comparison and merging of business processes.And BPG(Business process graph)is defined as a process description.BPG is a graphical representation of the existing modeling methods,it reduces some type of elements and only keep the elements describing the business behavior.Using BPG to model business process make it easy to analyze and process business process automatically by algorithms.On the other hand,BPG shields difference between different modeling languages by defining mapping rules from BPG to other modeling language.Based on BPG,the similarity metrics of business process is defined,including node similarity metrics and model similarity metrics.The node similarity metrics is used to match the node pairs in the models,which is mainly measured by label similarity and context similarity and sometimes supplemented by attribute and type similarity.The similarity of the models integrates the similarity of the nodes through a certain calculation model,it is used to measure the similarity between business models.Two model similarity metrics are defined,including node matching similarity and structure similarity.The former simply sums up the similarity values of matching node pairs.And the latter takes intoaccount the structural information of the graph and integrates the similarity values of node pairs based on graph editing distance.To calculate the similarity value of the models,it is necessary to establish the best matching between the models,that is,the mapping of nodes between models with the highest similarity score.In this paper,two heuristic model matching algorithms are proposed,the greedy strategy models matching algorithm and A* heuristic models matching algorithm.The former uses greedy strategy to select next pair of nodes to add to the matching set.The latter maintains a matching set,each time it selects the optimal matching and generates new matching to join the set.It improves the performance of algorithm by using the threshold cutoff to filter out the node pairs with low similarity when generating a new matching.Using the best matching as input,the algorithm to compute the intersections and complements between two models is defined.The algorithm identifies the common parts of the models by constructing the largest common area in the matching nodes,and the complements is the parts which are not in largest common area.Business process merging can not only facilitate the reusing of models,but also can achieves co-evolution of the input models and the merging model by optimizing the merging model and reflecting this optimization into the input models.So except the requirement of behavior retention,the model merging method should meet traceability and reversibility requirements.In order to achieve these goals,BPG has been improved to get configurable BPG that achieves traceability and reversibility by adding information about the source model of each elements in merging model.Based on configurable BPG,the model merging algorithm is given.And three kinds of model optimization methods are given to solve the redundancy and deficiency in the merge result.The effectiveness and performance of the business model comparison,merging and optimization methods were evaluated by the experiment which using SAP reference model as data set.The results show that the accuracy of structural similarity metrics is significantly higher than node matching similarity metrics in the experimental scenario of model retrieval.The accuracy of A* matching algorithm is higher than greedy matching algorithm,and greedy matching algorithm is more advantageous in execution time.In addition,the model merging and optimization algorithm also have satisfying results ineffectiveness and performance.