A declarative approach to customize workflow provenance

Abstract

Provenance describes the origin, context, derivation, and ownership of data products and is becoming increasingly important in scientific applications. This information can be used, e.g., to explain, debug, and reproduce the results of computational experiments, or to determine the validity and quality of data products. In contrast, it may be infeasible or undesirable to share complete provenance of a scientific experiment. Towards finding a balance between these requirements, we develop a framework and a system that allows scientists to declaratively specify their provenance data publication and customization requirements. Using this system, scientists can specify which parts of the provenance data are to be included in the result and which parts should be hidden, or anonymized. However, arbitrary application of these specifications may not maintain provenance data integrity. Thus, we allow scientists to specify provenance data integrity requirements, in form of provenance policies, along with their provenance data publication and customization requirements. Our system then systematically applies all the publication and customization requirements on the provenance data and ensures all the provenance policies as specified by the scientist.

Automated Phrase Mining from Massive Text Corpora

“…In contrast to keyphrase extraction [17], [27], [36], [24], [33], this task goes beyond the scope of single documents and utilizes useful cross-document signals. In [4], [14], [29], interesting phrases can be queried efficiently for ad-hoc subsets of a corpus, while the phrases are based on simple frequent pattern mining methods. The natural language processing (NLP) community has conducted extensive studies typically referred to as automatic term recognition [36], [13], [31], [38], [3], for the computational task of extracting terms (such as technical phrases).…”

Obscuring Provenance Confidential Information via Graph Transformation

“…A redaction-based graph grammar [13] for rewriting provenance graph replaces two or more nodes and the edges connecting them with a new node and applies node relabelling as necessary to hide sensitive information. The paper [14] shows how a variety of user requests such as abstracting, anonymizing, or hiding nodes may lead to provenance policy violations such as false dependencies, false independencies, or cyclic graphs. The paper suggests inventing new non-functional nodes when it is necessary and maintaining the essential relationships.…”