Abstract
Data integration typically relies on more than raw triples: provenance, quality indicators, usage policies,
and signatures are often required as in-band contextual statements. In the Resource Description Framework
(RDF), such annotations are expressed through a range of annotation models and method-specific practices.
In this space, mature systems such as DQV, nanopublications, RO-Crates, and W3C Verifiable Credentials
differ both in how they model annotations and how the annotation target is defined, with contextual
information encoded in the annotation system rather than at the data level. This heterogeneity limits the
uniform storage, exchange, discovery, and querying of contextual information associated with target
statements. We present Context Associations, an approach for uniformly modeling and querying associations
between contextual information and statements in an RDF knowledge graph. Our approach enables a lossless
and reversible conversion of existing annotations into a single association model based on blank-node
graphs. We evaluate Context Associations across the aforementioned annotation systems and show that
contextual information can be uniformly associated with target statements and queried across applications.
We further show that the original formats can be fully reconstructed when method-specific modeling
assumptions are made explicit. By providing a uniform representation of contextual information associated
with RDF statements, Context Associations supports the discovery, exchange, storage, and processing of
heterogeneous annotations.