Diego Rincon-Yanez, a researcher at the ADAPT Centre in Trinity College Dublin, has had his paper accepted to the main research track of the prestigious SEMANTiCS 2025 conference. The paper, titled “Document-Level Relation Extraction with Ontology-Guided RAG” tackles one of the more complex challenges in knowledge graph automation: Extracting relationships between entities across entire documents.
The work introduces GrOWL-RAG, a novel framework that combines Retrieval-Augmented Generation (RAG) with ontology-based reasoning to improve document-level relation extraction (docRE). Unlike traditional methods that rely on fine-tuning large language models, GrOWL-RAG enriches model prompts with structured knowledge from ontologies, enabling better context and precision.
Tested on the Re-DocRED benchmark, the framework showed promising results and outperformed baseline LLM approaches in ablation studies, highlighting the added value of ontology integration for this task.
This contribution highlights the increasing importance of Neuro Symbolic AI (hybrid AI) approaches in enhancing baseline tasks, such as Relation Extraction. This research was done in collaboration with Bosch AI Research with which the group has signed a collaboration agreement. For more information about the conference, visit the website here.
