Improve the capabilities of Large Language Models (LLMs like Gemini or ChatGPT) in interfacing RDF Data and RDF Knowledge Graphs. In order to do so as a major step, the aim is to build services that allow LLMs to lookup the concept in ontologies using DBpedia Archivo as flexible source for ontologies.
- improve an existing term search API based on DBpedia Lookup and DBpedia Archivo, such that it can be integrated via langchain/chatgpt/claude plugins to search for ontologies, classes, properties
- load the ontologies into a vector database such that LLMs can find relevant information in vector space
- integrate popularity information as mean to rank candidates (LOD stats, void stats generate e.g. from SPARQL endpoint)
- test and measure and compare the quality/performance (improvement) with the llm-kg-bench framework based by extending and adding evaluation test cases (converting a factsheet or csv into a KG, writing meaningful queries towards a SPARQL endpoint, mapping ontologies or datasets)
- 175 hours or 350 hours
Johannes Frey, Dr. Mahdi Hedayat Mahmoudi, Hannes Hartmann