Scientific Workflow Clustering And Recommendation Leveraging Layer Hierarchical Analysis

With the development of the Web technology,data and computational resources distributed among the Internet,like applications,are wrapped as Web services,such that they can be leveraged smoothly,ignoring different structure existing between different platforms.In order to construct a web service with new functionality,those existed Web services are composed as a new Web service.With a volume of discussion on the recommendation of service composition,service compositions are regarded as scientific workflow,and published on the public repository.Scientific workflow is composed of activity,sub-workflow and dependent links between them.Scientific workflows increase rapidly on the web,which facilitates workflow discovery,reuse and repurposing.In this setting,when scientists are going to develop a new experiment,construct a new workflow from scratch is laborious,which requires specific knowledge about possible activities,and the dependent links between them.Consequently,the reuse and repurposing of scientific workflow become a hot topic in this setting.Currently,some scientific research have worked on the semantic assessment to facilitate the workflow recommendation,which are categorize into two aspects: structure-based an annotation-based.However,layer hierarchy is ignored by most of them,which specify workflow with diverse granularity.To mitigate this problem,this thesis proposes a novel workflow recommendation technique.Specifically,1)Based on the relations between workflow,sub-workflow,and activity,this thesis proposes layer hierarchy to reconstruct scientific workflow,so as to emphasize diverse granularities for scientific workflow,2)Based on the tree edit distance and semantic assessment,this thesis presentsa semantic similarity method for scientific workflow considering layer hierarchy,including its hierarchical structure,tree-based workflow structure and semantic information,so as to construct a workflow network model for indicating similar relations among workflows,3)Based on the workflow network model,a graph clustering algorithm is proposed and leveraged for workflow classification.Core workflows are identified for representing corresponding cluster.An innovative layer hierarchy recommendation technology is presented for stimulating workflow reuse and repurposing.Finally,experiments are constructed to verify the efficiency and accuracy of the aforementioned workflow clustering and recommendation technique.Comparison experiments are constructed by using different clustering algorithm.This thesis presentsmetrics to evaluate experiment result.As a result,clustering algorithm proposed in this thesis is proven to be more efficient and accurate for retrieving candidate scientific workflows with respect to the requirement of scientists.