NutriLink: An Ontology for Linking Digital Receipts to Food Nutrition Information and Dietary Recommendations

REVISIONS REQUIRED:
1-"Food and Grocery Shopping Ontologies" requires a comparision table, too much similar info may confuse authors.
2- In Section "3. The NutriLink Ontology":
authors said "The NutriLink ontology was created following SAMOD [13], based on 3 key components:"
i) digital receipts from loyalty cards; --> how the researchers got these data, in which way?
ii) food product nutrition information from an existing FCD; --> How will the information of billions of products offered for sale under the same food category in market chains around the world be accessed and processed (offline/online)?
3- for section 3, authors may put a flow /model diagram to see the components used /required for the porposed system.
4-Digital receipts are not the same in every country around the world; only large grocery chains may apply the standard format, raising questions that cast doubt on the feasibility of the entire system.
5-Digital receipts may not always be reliable, and may involve product returns or misscanning issues. Seeing of these probs may create other usability or reliability problems.
6-Digital receipts reflect the dietary patterns of the entire family; it is not clear how they are reduced to the level of individuals.
7--Your work is comprehensive and extremely useful, and your research has reached a good conclusion. But I have some confusing questions about the real-life applicability and widespread impact of your work. The developed ontology includes some other ontologies and vocabularies to describe key features such as digital invoices, food products, nutritional information and dietary recommendations. It is very important to consider how this ontology will be used in practical application and how it will be integrated into supermarket chains and other food retail outlets. For example, integrating this ontology into a supermarket chain's digital infrastructure could enable customers to make more informed shopping decisions based on the health risks or nutritional values ​​they consume. Additionally, such a system has significant potential to provide recommendations to healthcare professionals, nutritionists and users. However, the practical challenges of market integration and adaptation need to be comprehensively addressed. Maybe you can add a short section on this topic.
8-Semnatic rules can be determined during the entegration stage or at the user level, and the necessary inferences can be provided by using SWRL on the developed ontology through these SWRL rules.
9-Some terms cannot be read on the user screens in Figure 2.
Please check these 3 useful and most related articles and if you find proper you can cite them:
a. FoodWiki: a mobile app examines side effects of food additives via semantic web. Journal of medical systems, 40(2), 41.
b. A safety food consumption mobile system through semantic web technology. In 2014 IEEE 38th International Computer Software and Applications Conference Workshops (pp. 348-353). IEEE.
c. A search service for food consumption mobile applications via hadoop and mapreduce technology. In 2016 IEEE 40th Annual Computer Software and Applications Conference (COMPSAC) (Vol. 2, pp. 77-82). IEEE.

NutriLink: An Ontology for Linking Digital Receipts to Food Nutrition Information and Dietary

I would like to commend the authors on their significant contributions research on semantics and data science more broadly, as well as to food and nutrition research. The creation of the NutriLink ontology represents an important advancement in linking digital receipts to nutritional information, which has the potential to provide actionable insights for guide healthier dietary behaviours, as well as a solid foundation for much future research. The integration of NutriLink with other ontologies like FoodOn and GoodRelations, along with its application in a diet recommendation system, is impressive and innovative. Overall, my comments below serve to improve the positioning of the paper, better explain the methods used, and address some other limitations.

Abstract:
- Likely needs to be reworked a little based on the comments below.
- Mention the country where the research was done, as its relevant in relation to nutrition guidelines as well as the retail data used on the case study.

Introduction:
- Most of the motivation for the project relies on legal reasoning (e.g., GDPR, EU regulation 1169/2011), but I am not sure if/how these regulations apply. My understanding was that nutrition information was mandated to be provided on food labels (front/back of pack), whereas I see the authors taking a more informatics lens to aggregate this data across products, shopping trips etc and provide more data than is legally required. Does this legal framework really necessitate providing this more comprehensive data or does it just apply to labelling?
- My understanding would be that retailers may be required to provide the transactional data, whereas the food manufactures are responsible for providing the nutritional content data of the food product. Note that fresh categories (meat, fruit & veg) are typically not required to have a nutrition facts label. Perhaps a brief description of responsibilities for data required from retailers vs. manufacturers vs. producers (farmers) would be helpful.
- The introduction may also benefit by motivating the research project by first discussing the more macro-level nutrition/health challenges facing society (e.g., high incidence of diet-related chronic diseases like diabetes, some cancers, obesity) and then discuss what tools/interventions have been tried before to address. Then maybe move to digital tools and how this project addresses a critical gap (which was started to be discussed on lines 5 to 8 of page 2 with the self-report vs. automated tracking).
- Also, linking nutrition information and digital receipts isn’t necessarily new. I know of several researchers who have used Neilsen IQ market data and linked it to food label information (for example, see Mary L’Abbé’s work at University of Toronto in Canada), or nutrition facts with loyalty card data already (see recent work by Mikael Fogelhom at University of Helsinki in Finland). Explaining what this project contributes beyond this existing work would be important.
- Page 2 lines 30-32 “this ontology also includes fine-grained Nutri-Score [12] details at the levels of both individual products and aggregated baskets.” Need to explain what Nutri-Score is to readers here and why it’s important. At this point the reader does not know what it is, especially for those in geographies where it has not been implemented.

Related work:
- Nutri-Score is just one of many nutrition scoring systems worldwide. There are also traffic light systems, warning labels (e.g., Canada’s new FOP labels; high in sugar/salt etc), guiding stars ratings, and others. This work should be reviewed briefly, and then explain why Nutri-Score was chosen for this project relative to others.
- Regarding the Food Composition Database (FCD) (lines 34-43 on page 3), more details are required regarding how the database was created/ where the data was obtained? Also, what is involved in maintaining the DB as mentioned on line 41?
- A discussion of recommendation systems is warranted. Broadly and within food retailing. What is currently done and what does this project contribute? I know most of the recommenders are based on past purchase history only, and not nutrition per se, but this should be discussed.
- Also, a discussion of what and how the nutrition recommendations are structured in Switzerland would be beneficial before moving to the Nutri-link ontology.

Nutri-link ontology:
- Page 5 lines 16-21. It is not clear how the Nutri-Score is aggregated at the basket level. Please explain because typically it is for an individual product.
- CQ4: Monthly energy and expense calculations… I think, at least, protein and fat are equally important macro-nutrients as energy. Also, maybe explain the difference between energy and how its calculated (vs. carbohydrates) somewhere.

3.2 Semantic model:
- Your list of re-used ontologies (page 5-6) is missing FoodON.
- Why is Dublin Core vocabulary needed? Do you also have images of the products in this ontology and the case study dataset?
- Page 6 lines 5-26. You have referenced how Nutri-link integrates with several different ontologies (FoodON, QUDT etc.) but there is no discussion of the methodology used to link these different ontologies. Was an NLP package used to match text between terms/definitions in the ontologies? What’s the degree of overlap and were there any challenges or missing links?

Case study:
- Page 9 line 21. 100 users at maximum is awkward. Do you have an accurate figure?
- Page 9 lines 30-33. In regard to “linking” the receipt data to the NutriLink, it is unclear how this matching is done. Again, are you using NLP to match item descriptions from the 2 retailers to the terms used in Nutrilink?
- Overall, I still a little confused about the dietary recommendations and the recommendation system. By recommendations, are you only referring to the text (e.g., reduce energy from sweet snacks) or are you also referring to the recommended “healthier products” that are depicted in Figure 2 [bottom, right screenshot]?

Limitations & Future work:
- One of the challenges of using receipt data is not necessarily knowing the household size and/or the occasion that the food was purchased for (e.g., for a party and not consuming all the food themselves). How has this been addressed in your project?
- Also, for the nutritional analysis section, consumers may buy food from other sources (e.g., restaurants, convenience stores, small shops), so how can the system be designed to incorporate this aspect of missing purchases? Or is it just a major limitation that needs to be acknowledged?
- Great to mention extension of this work to sustainability, but a bit more explanation would be nice. Sometimes sustainability refers to environmental footprint (e.g., CO2 emissions during production/processing), but it can also incorporate social aspects relating to equity (e.g., sustainable development goals). This also sometimes appears through food labels (e.g., free trade, B-corp certification, woman/Black/Indigenous-owned) How could social dimensions of sustainability be considered within this project in the future?
- For future research, I also think that there will be a large need for implementation science and designing of the human-computer interface (UX/UI) to promote uptake of the digital tools, as well as ensure that the data, insights, and recommendations actually produce a change in people’s food choices and nudge them in a healthier and more sustainable direction.
- Also, how can other marketing/promotional data beyond just prices (e.g., coupons, discounts, bonus points etc.) be incorporated in future work and perhaps in relation to the product recommendations?

Thank you again for this compelling and well-executed research. I look forward to seeing its continued development and application.

This paper presents NutriLink, an extensive ontology designed to bridge the gap between digital grocery receipts, food nutrition information and dietary guidelines. NutriLink provides basket- and product-level insights into dietary habits while also addressing gaps present within current ontologies, which form the structure for nutrition guidance. Its development, semantic model, integration with existing ontologies, and use in a fully automated diet counselling system are presented. The impact of this study may lie in combining personal shopping data with nutritional advice, a step toward personalized diet monitoring.
Following the journal recommendations we assess the following dimensions:
(1) Originality
The idea of adding nutrition and dietary recommendations to digital receipts while natural and reasonable, is new. Most existing ontologies such as FoodOn, GoodRelations and AGROVOC emphasize the specification of single food products or certain food domains but leave out the linking of receipts and dietary guidance on a more foundational level which NutriLink aims to fill. This is a major value addition to automated diet monitoring systems. A weakness to highlight is the relatively limited scope. i.e. supermarkets with a digital receipt system, but this is something that can be extended as information sources grow.
(2) Significance
The NutriLink ontology makes a promising step toward an effective enabling automatic, personalized diet recommendations. In nutrition informatics, its relevance and significance can be seen in a diet counselling system. The pilot study performed on 100 individuals seems rigorous and, while its effect cannot be checked, and as a consequence generalizability is still an issue, it seems, again, a step in the right direction.
3) Quality of Writing
The manuscript itself is well written, with clear definitions of technical terms and good organization. Sections on introduction and background give enough context. A final pass is needed to ensure grammatical accuracy and consistent use of terms, such as aligning "NutriScore" and "Nutri-Score" throughout the text. Also, some space for improvement can be found in more technical sections for example where SPARQL queries are described. These queries, while interesting from the technical point of view, my affect readability for non-technical readers (e.g. nutrition or health professionals) who are more interested on the ontology’s conceptual contributions. Most of these queries could be moved into supplementary materials or appendices and have organizational descriptions (summaries) remain in the main text for clarity and narrative flow.

4. The NutriLink ontology provides a permanent url though a GitHub repository.
The repository includes well-structured data files, with ontology documentation and SPARQL queries available. Besides this a README file is present, detailing the ontology's purpose, structure, and instructions for use. This makes the data accessible for replication.
The resources appear sufficient for replicating the experiments. The detailed competency questions and the SPARQL queries align with the ontology, facilitating reproducibility.

The manuscript presents a well-developed and innovative ontology, NutriLink, which addresses a critical gap in linking digital receipts with nutrition information and dietary recommendations. Its integration with established ontologies and practical application in a diet counselling system strengthens its potential impact in nutrition informatics. Overall, the study is a solid contribution to the field,
Minor Changes suggested.
Overall, the manuscript is original, well written and suggests what can be, in the middle term, a significant contribution. Some minor changes are suggested to contribute to its overall quality:
1. Improve the explanation of competency questions by providing brief, high-level descriptions of each competency question (CQ) in the text before diving into technical details. This will enhance readability for a broader audience.
2. Clarify the Nutri-Score updates, for example with a brief summary of the changes in the 2023 Nutri-Score framework and their implications for the NutriLink ontology.
3. Relocate technical SPARQL query examples to supplemental materials or appendices to streamline the manuscript for readers who are less familiar with semantic web technologies.
4. Include practical suggestions on how to adapt the ontology to non-supermarket datasets.
5. Enhance figure captions by providing more detailed captions for all figures, explaining their relevance to the ontology and system functionality.
6. A final review of the language is required to ensure grammatical accuracy and consistent use of terms.

7. Improve description of results from the pilot study (e.g., user engagement or preliminary feedback) to emphasize the system’s practical utility.

While it does not seem feasible nor required for this current publication, some suggestions for Future Work have arisen during the review process, that are indicated in the following.
1. Extend the ontology to include non-supermarket food items: Broaden the scope to cover restaurant meals, farmer’s market products, and home-prepared foods. This may make NutriLink more versatile in real-world applications.
2. Incorporate sustainability metrics, introducing concepts related to environmental sustainability, such as carbon footprint or water usage, to align with growing interest in sustainable consumption.
3. Conduct studies to test the ontology's applicability across different geographic regions, retail systems, and cultural dietary habits. This would really contribute to its generalizability.
4. Explore integrating real-time feedback mechanisms based on purchase behaviour, potentially using mobile apps or smart devices.
5. Include details about food preparation, cooking, and waste to provide a more accurate representation of nutritional intake.
6. Fully integrate the 2023 Nutri-Score updates and explore compatibility with alternative food labelling systems used outside Europe.
7. Enhance interactivity in diet counselling by developing personalized recommendation features that account for user preferences, health goals, and prior purchase history.
8. Develop training resources e.g. training materials or user guides for researchers and dieticians to facilitate broader adoption of NutriLink.