Agent-Based Provenance Architecture

An Agent-Based Provenance Architecture is described that supports the automated capture of provenance in a variety of domains, including physical and biological sciences, data ware-housing, workflow management systems, document management systems and geographic information systems. Disparate systems can store provenance of creation, ownership, transformation and other lineage using transparent, machine-understandable semantics that support data discovery, trust in the possible presence of deception and fitness-for-use determination. Resource Description Framework (RDF) triples are used for an open, extensible representation of provenance information. This Provenance Agent Architecture can apply to both legacy and future systems, fully automated systems, human-in-the-loop systems, stand-alone systems, fully distributed systems and client-server systems. After conceptualizing the agent architecture for data provenance, experiments were conducted to validate its key aspects, including the loose coupling of agents to applications and each other and applicability to multiple domains. This research leads to a society of agents that can handle provenance at differing levels of detail and can present provenance, at various levels of detail, usefully and intelligibly to end users. Support of this architecture for security and trust is described.