By MAIZE.
Explore how Large Language Models unlock structured knowledge from food safety data using AI and semantic resources.
This article presents how EFRA leverages instruction-tuned Large Language Models (LLMs) and retrieval-augmented techniques to automate knowledge extraction from food safety incident reports. By integrating external ontologies and a hierarchical classification system, the approach improves the scalability, accuracy, and interpretability of entity linking in complex label spaces.
Key Highlights:
Uses instruction-tuned LLMs for annotation without manual labels
Applies RAG-like architecture for entity linking in large taxonomies
Integrates multiple ontologies (AGROVOC, FoodOn, ChEBI, GS1 GPC)
Implements an LLM agent to automate classification workflows
Enhances scalability and accuracy in hierarchical classification
