Beyond Fine-Tuning: Robust Food Entity Linking under Ontology Drift with FoodOntoRAG
arXiv:2603.09758v1 Announce Type: new Abstract: Standardizing food terms from product labels and menus into ontology concepts is a prerequisite for trustworthy dietary assessment and safety reporting. The dominant approach to Named Entity Linking (NEL) in the food and nutrition domains fine-tunes Large Language Models (LLMs) on task-specific corpora. Although effective, fine-tuning incurs substantial computational cost, ties models to a particular ontology snapshot (i.e., version), and degrades under ontology drift. This paper presents FoodOntoRAG, a model- and ontology-agnostic pipeline that performs few-shot NEL by retrieving candidate entities from domain ontologies and conditioning an LLM on structured evidence (food labels, synonyms, definitions, and relations). A hybrid lexical--semantic retriever enumerates candidates; a selector agent chooses a best match with rationale; a separate scorer agent calibrates confidence; and, when confidence falls below a threshold, a synonym gene
arXiv:2603.09758v1 Announce Type: new Abstract: Standardizing food terms from product labels and menus into ontology concepts is a prerequisite for trustworthy dietary assessment and safety reporting. The dominant approach to Named Entity Linking (NEL) in the food and nutrition domains fine-tunes Large Language Models (LLMs) on task-specific corpora. Although effective, fine-tuning incurs substantial computational cost, ties models to a particular ontology snapshot (i.e., version), and degrades under ontology drift. This paper presents FoodOntoRAG, a model- and ontology-agnostic pipeline that performs few-shot NEL by retrieving candidate entities from domain ontologies and conditioning an LLM on structured evidence (food labels, synonyms, definitions, and relations). A hybrid lexical--semantic retriever enumerates candidates; a selector agent chooses a best match with rationale; a separate scorer agent calibrates confidence; and, when confidence falls below a threshold, a synonym generator agent proposes reformulations to re-enter the loop. The pipeline approaches state-of-the-art accuracy while revealing gaps and inconsistencies in existing annotations. The design avoids fine-tuning, improves robustness to ontology evolution, and yields interpretable decisions through grounded justifications.
Executive Summary
The paper presents FoodOntoRAG, a model- and ontology-agnostic pipeline for robust food entity linking under ontology drift. This innovative approach avoids fine-tuning, improves robustness to ontology evolution, and provides interpretable decisions through grounded justifications. By leveraging domain ontologies and structured evidence, FoodOntoRAG achieves state-of-the-art accuracy while revealing gaps and inconsistencies in existing annotations. The design has significant implications for trustworthy dietary assessment and safety reporting, particularly in the context of evolving food ontologies.
Key Points
- ▸ FoodOntoRAG is a model- and ontology-agnostic pipeline for robust food entity linking.
- ▸ The pipeline avoids fine-tuning and improves robustness to ontology evolution.
- ▸ FoodOntoRAG uses a hybrid lexical--semantic retriever and conditioning on structured evidence to achieve state-of-the-art accuracy.
Merits
Robustness to Ontology Evolution
FoodOntoRAG's design avoids the issue of fine-tuning on a particular ontology snapshot, making it more robust to ontology evolution and drift.
Interpretable Decisions
The pipeline provides interpretable decisions through grounded justifications, which can be particularly useful in high-stakes applications such as dietary assessment and safety reporting.
Demerits
Computational Complexity
While FoodOntoRAG's design avoids fine-tuning, it may still incur significant computational costs due to the use of multiple agents and retrievers.
Limited Evaluation
The paper's evaluation primarily focuses on accuracy and robustness, with limited exploration of other important metrics such as efficiency and scalability.
Expert Commentary
While FoodOntoRAG's design is highly innovative and effective, it is essential to consider the computational complexity and limited evaluation of the pipeline. Additionally, the paper's focus on accuracy and robustness raises important questions about the broader implications of ontology evolution and drift. Furthermore, the provision of grounded justifications highlights the importance of interpretable AI and explainability in high-stakes applications. Future work should aim to further investigate the pipeline's efficiency and scalability, as well as its applicability to other domains.
Recommendations
- ✓ Future research should explore the application of FoodOntoRAG to other domains and the development of more efficient and scalable versions of the pipeline.
- ✓ The paper's findings on ontology evolution and drift should be further explored in the context of policy and regulatory frameworks in the food and nutrition domain.