LabelFlow: Exploiting workflow provenance to surface scientific data provenance

Abstract

Provenance traces captured by scientific workflows can be useful for designing, debugging and maintenance. However, our experience suggests that they are of limited use for reporting results, in part because traces do not comprise domain-specific annotations needed for explaining results, and the black-box nature of some workflow activities. We show that by basic mark-up of the data processing within activities and using a set of domain specific label generation functions, standard workflow provenance can be utilised as a platform for the labelling of data artefacts. These labels can in turn aid selection of data subsets and proxy for data descriptors for shared datasets.