ProbSPARQL Extends Knowledge Graph Querying for Uncertain Numeric Data

The SFB 1574 Circular Factory project faces a significant challenge in integrating data about returned products into a shared knowledge graph, particularly because this data often includes multi-dimensional, uncertain numeric measurements from sensors. Existing RDF and SPARQL technologies lack native capabilities to effectively query and analyze such uncertain data, which is crucial for downstream processes like triage, validation, and reassembly planning. To bridge this gap, ProbSPARQL has been developed as an upward-compatible extension to SPARQL. This new framework models uncertain numeric values as random variables, with their distributions encoded using probabilistic RDF literal datatypes. ProbSPARQL supports advanced features such as distribution-aware expressions, probabilistic filters, and divergence-based joins. Implemented on Apache Jena ARQ and exposed via a Fuseki-compatible layer, its applicability was demonstrated using real-world measurement fragments and scalability was assessed on benchmarks up to 1.5 million triples, showing feasible in-engine execution and performance benefits over application-layer post-processing.