Reliable granular references to changing linked data

Abstract

Nanopublications are a concept to represent Linked Data in a granular and provenance-aware manner, which has been successfully applied to a number of scientific datasets. We demonstrated in previous work how we can establish reliable and verifiable identifiers for nanopublications and sets thereof. Further adoption of these techniques, however, was probably hindered by the fact that nanopublications can lead to an explosion in the number of triples due to auxiliary information about the structure of each nanopublication and repetitive provenance and metadata. We demonstrate here that this significant overhead disappears once we take the version history of nanopublication datasets into account, calculate incremental updates, and allow users to deal with the specific subsets they need. We show that the total size and overhead of evolving scientific datasets is reduced, and typical subsets that researchers use for their analyses can be referenced and retrieved efficiently with optimized precision, persistence, and reliability.