Exploiting Provenance to Make Sense of Automated Decisions in Scientific Workflows

Scientific workflows may include automated decision steps, for instance to accept/reject certain data products during the course of an in silico experiment, based on an assessment of their quality. The trustworthiness of these workflows can be enhanced by providing the users with a trace and explanation of the outcome of these decisions. In this paper we present a provenance model that is designed specifically to support this task. The model applies to a particular type of subworkflow that is compiled automatically from a high-level specification of user-defined, quality-based data acceptance criteria. The keys to the effectiveness of the approach are that (i) these sub-workflows follow a predictable pattern structure, (ii) the purpose of their component services is defined using an ontology of Information Quality concepts, and (iii) the conceptual model for provenance is consistent with the ontology structure.