AN ONTOLOGY ENRICHMENT FRAMEWORK USING RETRIEVAL-AUGMENTED LARGE LANGUAGE MODELS

ICLR 2026 Conference Submission13929 Authors

Keywords: Ontology enrichment, Multimodal large language model, Retrieval-augmented generation, Knowledge graph, 4D printing

TL;DR: Scalable Ontology Enrichment via Multimodal Large Language Models and Symbolic Reasoning Integrating Scientific Articles, Datasets, and Knowledge Graphs

Abstract: Ontology enrichment, understood as the process of extending and refining existing ontologies with new concepts, relations, and instances, has become a critical task for building robust and up-to-date knowledge bases. The exponential growth of scientific publications, datasets, and multimodal resources makes manual enrichment highly impractical, creating the need for automated or semi-automated approaches. In this work, we propose a framework that leverages multimodal large language models and retrieval-augmented generation to support ontology enrichment. Our method systematically extracts semantic knowledge units, aligns them with existing ontological structures, and generates interlinked triples, thereby enhancing both the coverage and the expressivity of the ontology. This framework addresses the knowledge acquisition bottleneck by enabling scalable integration of heterogeneous resources and fostering cross-domain semantic interoperability. To illustrate its effectiveness, we apply the framework to the domain of 4D printing, a rapidly evolving field at the intersection of materials science, manufacturing, and design. By incorporating knowledge about materials, properties, stimuli interactions, process parameters, and design strategies, the framework enriches a domain-specific ontology and supports innovation in the development of programmable and multifunctional structures.

Primary Area: learning on graphs and other geometries & topologies

Submission Number: 13929