Utilizing LLMs and ontologies to query educational knowledge graphs

Knowledge Graphs (KGs) provide knowledge and data in a structured format with content from various fields. But the access to the knowledge graphs is done by experienced users, that is, users who know the syntax of the SPARQL language and the KG vocabulary (either in RDF Schema or in OWL) in order to be able to ask questions to exploit the knowledge graphs. However, this requires a lot of time and effort for most of the users, which makes KGs inaccessible to a large number of users. Large Language Models (LLMs) that have appeared recently can provide an alternative way to query knowledge graphs without the need to learn SPARQL and/or know the schema and vocabulary of them, eliminating the time and effort that ordinary users need to spend in order to use them. In this article, we present some experiments and their results illustrating how ChatGPT can help ordinary users to generate SPARQL queries, without knowing SPARQL, to effectively use knowledge graphs and exploit their wealth of data. We experimented with ChatGPT to explore whether it can generate SPARQL queries based on user's natural language input and a given vocabulary (ontology) about an educational knowledge graph. To this end we have devised a specific prompt template. Results indicate that LLMs can indeed help in this direction, given the fact that they are prompted properly, using good English language. We have also discussed some practical lessons learned through this experiment.