HeriKGQA Achieves 99.0% Accuracy in Cultural Heritage Sensor QA
November 13, 2026
HeriKGQA uses a knowledge-graph approach to query sensor data by converting natural language into SPARQL and deterministic code. Testing on 500 questions across five heritage sites showed 99.0% accuracy, outperforming baseline LLMs by 18.2 points.
HOW THIS AFFECTS YOU
●
builderYou can improve LLM accuracy on sensor data by offloading computation to SPARQL and deterministic code.
●
researcherYou can use heritage ontologies like CIDOC-CRM to constrain LLM intent parsing for complex datasets.