Time travel for knowledge graphs: live queries over RDF change histories
For knowledge graph practitioners needing live temporal queries, this library provides a practical solution that avoids offline ingestion and supports concurrent updates, though it is incremental over existing approaches.
The paper presents the Time Agnostic Library, a Python library for live temporal SPARQL queries on any SPARQL-compliant triplestore, supporting all six temporal retrieval needs and concurrent updates. Evaluation on the BEAR-B benchmark shows sub-linear scaling in execution time and memory consumption, and it outperforms the closest live system R43ples across all query types.
Performing time-traversal queries on RDF datasets remains unsupported in the most extensive knowledge graphs. Existing solutions either require offline ingestion, which prevents concurrent querying and updating, or operate live but with limited query coverage or triplestore dependency. This article presents the Time Agnostic Library, a Python library for performing temporal SPARQL queries live on any SPARQL-compliant triplestore, supporting all six temporal retrieval needs identified in the literature and concurrent updates. The methodology builds on the OpenCitations Data Model (OCDM), which records provenance using the Provenance Ontology (PROV-O) and SPARQL UPDATE operations. The library supports version materialization, single-version and cross-version structured queries, delta materialization, and single-delta and cross-delta structured queries over multi-triple patterns. Evaluation on the BEAR-B benchmark shows sub-linear scaling in both execution time and memory consumption as the number of versions increases. While preprocessing-based systems such as OSTRICH achieve faster query times, they require offline ingestion and cannot handle concurrent data updates. Against R43ples, the closest live system in architecture, the Time Agnostic Library is faster across all query types.