Towards self-configuring Knowledge Graph Construction Pipelines using LLMs - A Case Study with RML

Marvin Hofer, Johannes Frey, Erhard Rahm

Keywords: Knowledge Graph Construction, LLM-KG-Engineering, Automated RML Mapping Generation

TL;DR: Assessing RML mapping generation via LLMs using IMDB data and DBpedia as target Ontology

Abstract:

This paper explores using large language models (LLMs) to generate resource mapping language (RML) files in the RDF turtle format as a key step towards self-configuring RDF knowledge graph construction pipelines.
Our case study involves mapping a subset of the Internet Movie Database (IMDB) in JSON format given a target Movie ontology (selection of DBpedia Ontology OWL statements).
We define and compute several scores to assess both the generated mapping files and the resulting graph using a manually created reference.
Our findings demonstrate the promising potential of the state-of-the-art commercial LLMs in a zero-shot scenario.

Submission Number: 8