Reproduce “OntoExtend: A Framework for Requirement-driven and Scalable Ontology Extension with LLMs” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Anna Sofia Lippolis, Mohammad Javad Saeedizade, Stefan Schmid, Simon Blattner, Robin Keskisärkkä, Aldo Gangemi, Eva Blomqvist, Andrea Giovanni Nuzzolese

The OntoExtend framework, developed by researchers from Linköping University, University of Bologna, Bosch, and ISTC-CNR, presents a retrieval-augmented generation (RAG) approach for requirement-driven and scalable ontology extension using LLMs. It grounds LLM-generated extensions in retrieved relevant ontology fragments, resulting in high-quality outputs that correctly model all competency questions with less than 2% superfluous elements and require minor to moderate human revision.

Key insights

  • A RAG-based approach effectively mitigates LLM context window limitations and promotes reuse of existing ontology elements, enabling scalable and requirement-driven ontology extension.
  • The quality of LLM-generated ontology extensions is highly dependent on the precision and clarity of the input competency questions (CQs), with more specific CQs yielding superior results.
  • A multi-dimensional evaluation, combining automated structural checks (syntax, OOPS!, reasoner consistency) with functional assessment (CQ verification, superfluous elements) and expert human review, is crucial for robust assessment of LLM-generated ontology fragments.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.