Abstract
The deployment of various networks (e.g., Internet of Things [IoT] and mobile networks), databases (e.g., nutrition tables and food compositional databases), and social media (e.g., Instagram and Twitter) generates huge amounts of food data, which present researchers with an unprecedented opportunity to study various problems and applications in food science and industry via data-driven computational methods. However, these multi-source heterogeneous food data appear as information silos, leading to difficulty in fully exploiting these food data. The knowledge graph provides a unified and standardized conceptual terminology in a structured form, and thus can effectively organize these food data to benefit various applications. In this review, we provide a brief introduction to knowledge graphs and the evolution of food knowledge organization mainly from food ontology to food knowledge graphs. We then summarize seven representative applications of food knowledge graphs, such as new recipe development, diet-disease correlation discovery, and personalized dietary recommendation. We also discuss future directions in this field, such as multimodal food knowledge graph construction and food knowledge graphs for human health.
Introduction
The modern food industry generates a staggering amount of data, ranging from agricultural IoT sensors and nutritional databases to social media "food porn" and clinical health records. However, this wealth of information currently exists in "information silos." A recipe shared on Instagram may use entirely different naming conventions than a USDA nutrition table, which in turn differs from a hospital’s dietary log. This heterogeneity makes it nearly impossible for artificial intelligence to perform complex reasoning across datasets.
To address this, researchers are turning to Knowledge Graphs (KGs) as the missing link. By providing a unified, standardized framework to organize multidisciplinary data, KGs enable the "Internet of Food" (IoF)—a machine-readable ecosystem where data points from farm to fork can finally "talk" to each other.
Technical Foundations of the Internet of Food
Building a Food Knowledge Graph requires a clear understanding of Semantic Web technologies. A common point of confusion in the field is the distinction between an ontology and a knowledge graph. In essence, an ontology serves as the skeleton—defining classes like "Fruit" or "Nutrient"—while a Knowledge Graph is the flesh, containing millions of specific instances, their chemical compounds, and their roles in specific recipes.
To establish a common language, the following table outlines the core terminology used in the construction of these systems:
These technologies, such as RDF (Resource Description Framework) and OWL (Web Ontology Language), allow for the creation of Subject-Predicate-Object triples (e.g., "Apple—is a—Fruit") that form the basis of the graph structure.
Construction and Reasoning Methodologies
The transition from static databases to dynamic KGs involves a sophisticated pipeline:
- Construction: Modern systems are moving from manual expert curation toward auto-construction. This involves using Natural Language Processing (NLP) and machine learning to extract entities and relations from unstructured text like scientific papers or recipe blogs.
- Representation & Reasoning: By utilizing Knowledge Graph Embeddings, entities are transformed into vectors in a continuous mathematical space. This allows for "link prediction"—the ability to infer a relationship between a specific food and a disease even if that relationship was never explicitly stated in the source data.
- Advanced Architectures: The field is increasingly adopting Graph Neural Networks (GNNs). Unlike traditional AI, GNNs propagate information across neighboring nodes, allowing for more complex logical reasoning that mimics human-like contextual understanding.
Key Applications in the Food Domain
The review identifies seven primary areas where Food KGs are currently making an impact:
- New Recipe Development: KGs help identify healthy ingredient substitutions (e.g., replacing sugar with functional alternatives) and discover "latent ingredient combinations"—mathematically sound flavor pairings based on shared chemical compounds.
- Question-Answering (QA): Systems like FoodKG allow users to query complex nutritional needs, such as finding meals that are high in protein but low in sodium.
- Diet-Disease Discovery: Tools like NutriChem map associations between phytochemicals and diseases, aiding in preventative medicine.
- Knowledge-Augmented Vision: KGs act as "prior knowledge" for computer vision. If a model identifies noodles and peanuts, the KG suggests a high probability of "Pad Thai," boosting recognition accuracy by nearly 10%.
- Personalized Recommendation: By linking a user’s personal health data with a general Food KG, systems can provide clinically relevant, personalized dietary advice.
- Supply Chain & Manufacturing: KGs enable cross-border traceability of contaminated food and allow smart appliances to "recognize" and correctly cook specific food items.
Future Directions
Despite significant progress, several frontiers remain. Most current KGs are text-heavy; the next generation must be truly multimodal, integrating images, videos, and sensor data as first-class citizens. Furthermore, resolving contradictions between different cultural health beliefs and ensuring KGs can adapt dynamically to seasonal ingredient changes are major hurdles. Finally, deeper integration with human "gut microbiota" and genomic data will be essential for the future of precision nutrition.
Summary updated