Text-to-SPARQL for Proprietary Ontologies

This thesis explores the state of the art in text-to-query research, focusing specifically on text-to-SPARQL for proprietary ontologies rather than open knowledge graphs like Wikidata. The objective is to review existing methodologies and develop a prototype that translates natural language queries into SPARQL while optimizing context window usage.

A key challenge in this area is efficiently handling large ontologies without preloading the entire structure. This research aims to design an innovative mechanism that dynamically identifies and retrieves only the relevant ontology elements required for each query. Instead of relying on a static, full-context approach, the system will integrate cutting-edge techniques such as query decomposition and prompt engineering to enhance efficiency and accuracy.

While some commercial solutions already employ similar strategies, their implementations remain unpublished. This thesis will contribute by developing an open, modular approach that improves dynamic ontology extraction and advances the practical application of text-to-SPARQL systems.

This thesis is conducted in cooperation with Adorsys. If you are interested in this topic, please contact Mark Kram (mark.kram@adorsys.com).