Workshop on LLM-driven Knowledge Graph and Ontology Engineering (llms4kgoe) 2026

First Workshop | Co-located with ESWC 2026 (23rd European Semantic Web Conference)

About

The maturing of Semantic Web technologies cannot be separated from the need for high-quality, reusable ontologies. However, traditional ontology engineering is a notoriously difficult, time-consuming, and expert-driven process. The advent of LLMs presents a paradigm shift, promising to accelerate and democratize this process, enabling "specialists from beyond computer science" to develop their own models.

While the potential is enormous, automated ontology creation from LLMs presents significant challenges that the Semantic Web and knowledge engineering communities have yet to systematically address. If the foundational ontologies generated by LLMs are of poor quality, incoherent, or formally incorrect, the downstream AI systems built upon them will inherit these critical flaws.

This workshop focuses on the emerging and rapidly evolving intersection of Large Language Models (LLMs), semantic technologies, and ontology engineering. It addresses the challenges and opportunities associated with leveraging LLMs for ontology creation, refinement, and validation. The workshop encompasses both theoretical and practical aspects of LLM-based ontology construction, including semi-automated/fully automated ontology generation pipelines, evaluation methodologies, and strategies for reducing hallucinations.

Motivation: This workshop is motivated by three key observations: (1) the growing practical deployment of LLMs in knowledge graph and ontology construction projects with minimal formal evaluation frameworks; (2) the lack of systematic comparison between LLM-based and traditional ontology engineering approaches; and (3) the absence of community consensus on appropriate evaluation methodologies and quality metrics specific to LLM-generated ontologies.

Topics of Interest

We welcome contributions on topics concerning the development and assessment of high-quality ontologies, both manually engineered and automatically generated using large language models (LLMs). The main topics of interest include:

LLM-to-KG with schema constraints Enforcing structured templates and ontology schemas during triple generation.
Education & UX Developing LLM-driven tutors for CQ/axiom authoring and automated documentation.
Evidence-linked triple extraction Capturing direct evidence sentences and document sources for traceability.
Hallucination benchmarking Metrics and datasets for measuring hallucination severity in KG extraction.
Post-hoc KG repair Applying symbolic reasoners and neural consistency models to detect/correct errors.
Calibration and abstention Incorporating probabilistic calibration to allow abstention on uncertain links.
Robustness and red-teaming Stress-testing model robustness using adversarial inputs and perturbations.
Domain applications Deploying KG within domains (e.g., biomedical, climate) to quantify decision impact.
Lifecycle and Maintenance LLM-assisted ontology evolution, versioning, CI/CD integration, and refactoring.
Modular Evaluation & Benchmarks Component-level metrics, task cards, and error taxonomies.
Provenance & Governance Designing evidence-traceable axioms, audit trails, and governance mechanisms.
Neuro-symbolic Control Reasoner-in-the-loop decoding and learned validators for logical soundness.
Human-in-the-Loop Protocols Role hierarchies and economic evaluation for expert participation.
Domain Adaptation Adapters, RAG with domain-specific KGs, and drift management.
Multilingual & Multimodal OE Pipelines translating text, tables, and figures into consistent axioms.
Operational Efficiency Measuring cost, energy efficiency, and latency of proprietary vs. open-source models.
Standards & Community Alignment Integration with FAIR principles, OBO/ODP standards, OAEI, and LLMS4OL.

Workshop Format

Type: First Workshop
Length: Full Day (6-7 hours including breaks)
Mode: In-person (aligned with conference on-site infrastructure)

We plan the workshop to begin with an invited talk followed by presentations of contributed papers (with discussion), and eventually, by an interactive session of breakout groups working on and discussing ontology design issues and patterns. We will conclude with joint reporting and a community town hall.

Submission Instructions

The papers will be (single-blind) peer-reviewed. The workshop proceedings will be published by CEUR. We will also provide the option of not archiving the submissions on CEUR to the authors.

(Detailed submission links and deadlines to be updated)

Important Dates

Tentative Schedule

The following is a placeholder schedule based on typical workshop structures. Specific papers and times will be announced closer to the event.

Organization

General Chairs

Program Committee