Abstract
As artificial intelligence increasingly permeates food science—from calorie prediction to personalized nutrition—the field of Food Information Engineering has experienced rapid growth. However, this expansion has resulted in a fragmented research landscape where contributions are scattered across diverse journals and conferences. This paper presents a systematic approach to organizing this domain using the Open Research Knowledge Graph (ORKG). By structuring the contributions of over 120 papers into machine-readable templates, this work provides a centralized, evolving roadmap for researchers and engineers.
Introduction
Food Information Engineering encompasses the acquisition, processing, and dissemination of food-related data. While the integration of AI into kitchens and clinics offers significant potential, the "fragmentation problem" poses a substantial barrier to progress. New researchers often face a steep learning curve, and engineers struggle to identify state-of-the-art tools for tasks such as food image recognition or recommendation systems.
To address this, we have developed a systematic, accessible body of knowledge that moves beyond traditional static literature reviews. By leveraging structured data, we aim to bridge the gap between disparate research efforts and provide a unified framework for the community.
Methodology
The core of this research utilizes the Open Research Knowledge Graph (ORKG), a platform designed to transition scientific publishing from static PDFs toward structured, machine-interoperable data. The methodology was executed in two distinct phases:
- Exploratory Phase: A six-month curation and training period conducted in collaboration with the ORKG team. During this phase, 10 specific "templates" were developed. These templates serve as standardized forms to describe a paper’s contributions, capturing critical metadata such as datasets used, algorithms applied, and specific research outcomes.
- Systematic Literature Review (SLR) Phase: These templates were applied to systematically compare papers and populate the knowledge graph. This process is inherently iterative; as new methodologies emerge in the literature, the templates are refined to capture increasingly nuanced data points.
Data Synthesis and Infrastructure
While this study does not introduce a new benchmark dataset for model training, it establishes a "meta-dataset" of the existing literature. The corpus consists of 120 scientific papers, from which over 400 individual contributions were extracted.
The resulting infrastructure includes:
- 33 Comparison Tables: These allow users to filter research by data source (structured vs. unstructured), extraction methods, or specific applications like calorie prediction.
- 2 SmartReviews: Living literature reviews hosted on the ORKG platform that provide a narrative synthesis of the structured data.
- Standardized Templates: Five specific templates were dedicated to food composition tables and knowledge graphs, enabling "apples-to-apples" comparisons across different research groups.
Key Insights and Trends
Our analysis of the structured data revealed several critical trends in the evolution of Food Information Engineering:
Methodological Evolution
The field is currently undergoing a transition from manual, rule-based knowledge extraction to more sophisticated deep learning and Natural Language Processing (NLP) techniques. This shift is particularly evident in the construction of food ontologies and automated data harvesting.
The KG-to-QA Pipeline
A dominant architectural pattern identified in the literature is the use of Food Knowledge Graphs (KGs) as the foundation for Question Answering (QA) systems. This suggests that structured data remains the preferred backbone for building "expert" food AI systems that require high levels of accuracy and explainability.
The Neurosymbolic Gap
A significant finding of this review is the near-total absence of Neurosymbolic AI—the combination of neural networks with logic-based AI—within food information engineering. Given the success of this approach in other domains, its absence represents a notable research gap and a high-potential area for future exploration.
Conclusion and Future Work
This work establishes a public, digital infrastructure for Food Information Engineering, accessible at https://orkg.org/author/R138055. By moving research data into a machine-readable format, we provide a foundation for more efficient discovery and comparison of scientific contributions.
The systematic literature review remains an ongoing effort. Future work will focus on expanding the corpus of papers and refining the ORKG resources to keep pace with the rapidly evolving AI landscape. Specifically, we encourage the community to explore Neurosymbolic AI as a means to bridge the gap between raw data processing and logical reasoning in food science.
Summary updated