LLMs (Large Language Models), such as ChatGPT, are technologies with unprecedented text analysis capabilities. Despite the performance of LLMs is continuously improving, they still suffer from hallucinations and somehow limited capabilities when performing complex reasoning tasks. To explain LLMs results and perform complex reasoning tasks in specific domains, it is possible to adopt an Ontology. In general, Ontologies represent knowledge in a standard form using description logic. An ontology includes the main concepts and the possible relationships among them. Well established reasoning software can exploit this logic to perform complex deductions and check if statements are logically correct according to the ontology. This process can be used to verify the coherence of LLMs responses. Unfortunately building an ontology for a specific domain and populating it with relevant concepts and relationships is a labour-intensive task for humans.
This research project wants to investigate the idea to use LLM technologies (i.e., prompt engineering) to automatically build and populate an ontology for a specific domain (e.g., financial news, sport news etc.). Starting with a minimal set of concepts the approach will perform an Ontology Learning task looking for a cost-effective and scalable solution for knowledge acquisition. This will enable trusted and verified decision-making support in different domains exploiting LLM. The objective of the project is to design and implement a novel technique for Ontology Learning with LLMs to improve their effectiveness. Rigorous experiments will need to be designed and conducted to evaluate the effectiveness of the proposed approach.
Undergraduate
None
HASEL (405.662, Lab)