LLMs, Ontologies, and Knowledge Graph


The Large-Language Models, Ontologies, and Knowledge Graphs Working Group (LOK) is a collaborative effort that aims to bridge the gap between artificial intelligence (AI), specifically large language models, ontologies, and knowledge graphs. The group is composed of diverse members, including researchers from the National Center for Autonomous Systems Research, University of Buffalo, graduate students, international participants, and government contractors.

The working group operates under the belief that we are at a pivotal point where advances in generative AI and knowledge graph development can mutually benefit each other. The group’s primary focus is to leverage knowledge graphs to enhance large language models, identify biases, and create guardrails on their outputs. They also explore other ways to enhance knowledge graphs with large language models, such as construction, co-reference resolution, and Q&A entity and relation extraction.

The group encourages contributions from various fields, including literature reviews, collaborating on articles, testing of large language models and ontologies, and producing novel tools and methods. They plan bi-weekly meetings, with a focus on landscape review and information exchange in the first three months, and advancing current research through developing generative AI ontologies and co-authoring journal articles in the following months.

Despite the challenges in applying knowledge graphs in real-world settings due to the prevalence of non-graph-based data, the group remains optimistic about the potential of this technology to address concerns and improve the quality of generative AI. They invite questions, clarifications, and are open to new ideas and collaborations.

Scope and Purpose

The purpose of this working group is to explore ways in which knowledge representation research may complement and benefit large-language model research, and vice versa. Topics within scope of the working group include tools, strategies, design patterns, and benchmarks at the intersection of these fields.

Expected outcome of the WG

Initial deliverables include a meta-review on the state of convergence between LLM and knowledge representation research. A subsequent deliverable will include a set of benchmark tests to be housed in a repository for the evaluation of the utility of augmenting LLM implementations with knowledge representation research, and vice versa. A subsequent deliverable will be an empirically based demonstration of value add by leveraging LLMs/KGs in an operational environment.

Tentative schedule

This group is anticipated to meet biweekly, to be open-ended as the respective fields advance, but to return an initial ‘state of the art’ white paper within 6 months of the group initiation.

Member relationship

Collaborative commitment on an initial literature review. Evaluation and construction of practical benchmarks. Collaborative writing and discussion around empirical research demonstrating LLM/KG synergy Collaboration on writing articles and providing content for repositories.

Active members of this working group will enjoy the following rewards:

Credit as co-author on published papers; credit as co-creator of benchmarks; advertised in talks on topics covered and advertised as a contributor across network of connections (I like to credit where it is due)


Introductory Video

Join the discussion at:


John Beverley

Associate Professor, Norwegian University of Life Sciences