Crossing-workflow Fragments Discovery Leveraging Activity Knowledge Graph

At present,there is a large and increasing number of scientific workflows publicly accessible on the repository.Scientific workflows correspond to processes which should be executed recurrently by scientists to implement certain functionalities They are dependent on the activities,functional relations between activities,and sub-workflows.When scientists want to design and carry out new experiments,considering the fact that developing a novel scientific workflow from scratch is typically a knowledge-and effort-intensive,and error-prone mission,reusing and repurposing best-practices that have been evidenced by legacy scientific workflows is considered as a cost-effective and error-avoiding strategy.However,a scientific workflow experiment may be related to one or more scientific workflows.This observation drives us to propose a technology that can discover and recommend relevant crossing-workflow fragments to meet such needs.Our previous work has realized how to do the discovery and recommendation of crossing-workflow fragments for the requirement fragments composed of activities at the same level of granularity,but the constraint lies in the fact that except flat invocation relations,other relations like hierarchical parent-child relations may exist between workflows,their sub-workflows,and activities.A certain requirement may be satisfied by composing fragments with coarse or fine granularities with respect to the granular level of requirement specification.To remedy this issue,this paper proposes a novel crossing-workflow fragments discovery mechanism.The contributions of this paper are as follows:1)Leveraging legacy scientific workflows,an activity knowledge graph is constructed to capture flat invocation relations between activities in scientific workflows in the my Experiment repository and the hierarchical parent-child relations specified upon sub-workflows with their corresponding activities.Generally speaking,the activity knowledge graph can promote the discovery work of fragments;2)The semantic relevance of activities and sub-workflows is calculated based on their representative topics.The names and descriptions of activities and sub-workflows are converted into corresponding short documents,and topic models are applied to obtain the topic distribution on the activities and sub-workflows in the short document's corpus.According to a certain filtering strategy,representative topics are selected relative to each activity and sub-workflow;3)Given a requirement specified according to the workflow template,consider the semantic relevance and short document description,and find candidate activities or sub-workflows for each activity stub on the requirement template.Candidate fragments are constructed by discovering relations on candidate activity or sub-workflow specified activity knowledge graphs.The structural similarity and semantic similarity of the fragments are balanced,and these candidate fragments are evaluated and ranked.A large number of extensible experiments are performed using the scientific workflows data.Results demonstrate that our approach is accurate in discovering appropriate crossing-workflow fragments in comparison with the state of art's techniques.