Integrating SHAP/LIME and Knowledge Graph Reasoning to Explain Food and Menu Recommendations in EXERSYS
EXERSYS
An EXplainable RecommandER SYStem for the Nutrition Domain.
Combining Knowledge Graphs, Ontologies and Machine Learning.
Most chronic diseases are correlated to unhealthy eating habits [1]. Public health agencies have created dietary guidelines targeting the general population to push people for healthier eating habits: “eat at least 5 fruits or vegetables per day”, “limit your consumption of salt”. The compliance of these guidelines is relatively low, although the awareness about an healthy diets is rather good [2]. There are different causes that contribute to this: cultural and personal preferences, difficulty of implementation, availability and price of food items [3] and so on.
In the EXERSYS project we are developing a recommender system of food items that can deal with most of these causes. Recommender systems [4] are (web, mobile, standalone) tools that have become increasingly popular for supporting the user in finding personalized suggestions of products, services and information. They have been very successful in a variety of domains (e.g., movies, shopping, social networks) and deployed in many applications. Recommender systems (refer to [5] for a survey) are based on the general idea of “suggesting similar items for similar users” and often exploit personal user preferences, past behavior and similarity between users.
In food related recommender systems, the recommended objects can be recipes, food items or menus. A menu is a complex item composed of different dishes, users’ preferences for a dish can change in response to the other dishes consumed with it, users' health situation (e.g diabetes, arterial tension, allergies) may add constraints on possible dishes/ingredients to consider in a menu. Hence, recommending menus requires checking if the plates are compatible and fitting the user preferences and her health constraints. Moreover, a food-related recommender may consider the sequential aspect of the eating consumption (what we accept to eat today may be related to what we have eaten yesterday), while recommending an item to buy on a website is a one-shot recommendation, recommending a food-related item one needs to consider at which frequency the item should/could be recommended and when it has been consumed by the user. A food-related recommendation must also consider the context of the consumption: user's preferences for food-related items may also be dependent on user's context that can be social (e.g., dinner with friends), geographical and seasonal (e.g., recommending menus with seasonal ingredients). Finally, when knowledge and/or data are available, other constraints may be interesting to include such as ecological and ethical aspects that may concern the origin of the ingredients, their environmental impact (e.g., use of phytosanitary products, deforestation) and if their production is ethics compatible.
Menus recommendation is a novel challenging topic; recent works [18],[19] have dealt with the problem of recommending a menu but they do not consider the complexity of the problem such as the context of the consumption or the past sequence of menus consumed. The recommender system developed in this project tackles these challenges by considering a hybrid approach that uses knowledge graphs as prior knowledge and contextual information for the recommender system [Noemie][Alexandre].
One of the reasons that limit the acceptability of a recommendation is the absence of explanations behind it. The main focus of this master internship is to develop methods to provide such an explication based on the knowledge graph used by the recommender system.
The student will:
Conduct a literature review on the state of the art in explainability for recommender systems, with a particular focus on knowledge-based recommender systems.
Familiarize themselves with the current architecture and model of the existing recommender system.
Implement an explainability method following the current research in machine learning (e.g., SHAP, LIME, etc.).
Investigate approaches to enhance or complement the system’s explainability through the integration of the Knowledge Graph (KG).
Implement and evaluate the newly designed recommender system using existing datasets (INCA) and knowledge graphs (NutriKG).
Required technical skills:
Solid background in machine learning
Good understanding of knowledge representation and knowledge graphs (RDF, OWL, SPARQL).
Experience with Python and common data science libraries (e.g., scikit-learn, pandas, PyTorch, TensorFlow).
Basic knowledge of evaluation metrics for recommender systems (e.g., precision, recall, coverage, diversity).
Supervision team:
Cristina Manfredotti, AgroParisTech, UMR MIA Paris-Saclay 518
Fatiha Saïs, LISN, UMR CNRS 9015, Université Paris-Saclay
Stéphane Dervaux, INRAE, UMR MIA Paris-Saclay
Planned internship period: March 1, 2026 – August 31, 2026
Location: Palaiseau
Allowance: Approximately €650/month (expected to increase in 2026, final rate to be confirmed in November).
Bibliography:
[1] World health organization: diet, nutrition and the prevention of chronic diseases: report of a joint who/fao expert consultation, 2003.
[2] Smith Edge M. Ivens, B. J. Translating the dietary guidelines to promote behavior change: Perspectives from the food and nutrition science solutions joint task force. J Acad Nutr Diet, 116(10):1697–1702, 2016
[3] Byrd-Bredbenner C. Webb, D. Overcoming consumer inertia to dietary guidance. Advances in Nutrition, 6(4):391–396, 2015
[4] Gediminas Adomavicius, Bamshad Mobasher, Francesco Ricci, and Alexander Tuzhilin. Context-aware recommender systems. AI Mag., 32(3):67–80, 2011.
[5] Francesco Ricci, Lior Rokach, and Bracha Shapira, editors. Recommender Systems Handbook. Springer, 2015
[6] Badrul Munir Sarwar, George Karypis, Joseph A. Konstan, and John Riedl. Item-based collaborative filtering recommendation algorithms. In Vincent Y. Shen, Nobuo Saito, Michael R. Lyu, and Mary Ellen Zurko, editors, Proceedings of the Tenth International World Wide Web Conference, WWW 10, Hong Kong, China, May 1-5, 2001, pages 285–295. ACM, 2001
[7] Paul Sheridan, Mikael Onsjo, Claudia Jeanneth Becerra, Sergio Jimenez, and George Duenas. An ontology-based recommender system with an application to the star trek television franchise. Future Internet, 11(9):182, 2019.
[8] Zehuda Koren, Robert M. Bell, and Chris Volinsky. Matrix factorization techniques for recommender systems. Computer, 42(8):30–37, 2009
[9] Linas Baltrunas, Bernd Ludwig, and Francesco Ricci. Matrix factorization techniques for context aware recommendation. In Bamshad Mobasher, Robin D. Burke, Dietmar Jannach, and Gediminas Adomavicius, editors, Proceedings of the 2011 ACM Conference on Recommender Systems, RecSys 2011, Chicago, IL, USA, October 23-27, 2011, pages 301–304. ACM, 2011.
[10] Julien Delporte, Stephane Canu, and Alexandros Karatzoglou. Apprentissage et factorisation pour la recommandation. In Younes Bennani and Emmanuel Viennet, editors, Apprentissage Artificiel et Fouille de Données, AAFD 2012, Université Paris 13, Institut Galilée, Villetaneuse, France, 28-29 juin 2012, volume A-6 of RNTI, pages 1–26. Hermann-Editions, 2012
[11] Hong-Jian Xue, Xin-Yu Dai, Jianbing Zhang, Shujian Huang, and Jiajun Chen. Deep matrix factorization models for recommender systems. In Proceedings of the 26th International Joint Conference on Artificial Intelligence, IJCAI’17, page 3203–3209. AAAI Press, 2017
[12] Qingyu Guo, Fuzhen Zhuang, Chuan Qin, Hengshu Zhu, Xing Xie, Hui Xiong, and Qing He. A survey on knowledge graph-based recommender systems, 2020.
[13] Nicola Guarino, Daniel Oberle, and Steffen Staab. What is an ontology? In Steffen Staab and Rudi Studer, editors, Handbook on Ontologies, International Handbooks on Information Systems, pages 1–17. Springer Berlin Heidelberg, 2009.6
[14] Andrea Scharnhorst and Richard P. Smiraglia. Chapter 1. the need for knowledge organization. Introduction to the book linking knowledge: Linked open data for knowledge organization. In Linking Knowledge, pages 1–23. Ergon – ein Verlag in der Nomos Verlagsgesellschaft, 2021
[15] Franz Baader, Ian Horrocks, Carsten Lutz, and Ulrike Sattler. An Introduction to Description Logic. Cambridge University Press, 2017
[16] Qingyu Guo, Fuzhen Zhuang, Chuan Qin, Hengshu Zhu, Xing Xie, Hui Xiong, and Qing He. A survey on knowledge graph-based recommender systems. CoRR,abs/2003.00911, 2020
[17] Zongfeng Zhang and Xu Chen. Explainable recommendation: A survey and new perspectives. CoRR, abs/1804.11192, 2018. 8
[18] Imam Cholissodin and Ratih Kartika Dewi. Optimization of healthy diet menu variation using pso-sa. Journal of Information Technology and Computer Science, 2(1):28–40, Jun. 2017
[19] Massimo Quadrana, Paolo Cremonesi, and Dietmar Jannach. Sequence-aware recommender systems. ACM Comput. Surv., 51(4):66:1–66:36, 2018
[20] Melanie Munch, Juliette Dibie, Pierre-Henri Wuillemin, and Cristina E. Manfredotti. Towards interactive causal relation discovery driven by an ontology. In Roman Bartak and Keith W. Brawner, editors, Proceedings of the Thirty-Second International Florida Artificial Intelligence Research Society Conference, Sarasota, Florida, USA, May 19-22 2019, pages 504–508. AAAI Press, 2019
[21] Joe Raad, Nathalie Pernelle, and Fatiha Saıs. Detection of contextual identity links in a knowledge base. In ́Oscar Corcho, Krzysztof Janowicz, Giuseppe Rizzo, Ilaria Tiddi, and Daniel Garijo, editors, Proceedings of the Knowledge Capture Conference, K-CAP 2017, Austin, TX, USA, December 4-6, 2017, pages 8:1–8:8. ACM, 2017.
[22] Alina Petrova, Egor V. Kostylev, Bernardo Cuenca Grau, and Ian Horrocks. Query-based entity comparison in knowledge graphs revisited. In Chiara Ghidini, Olaf Hatig, Maria Maleshkova, Vojtech Svatek, Isabel Cruz, Aidan Hogan, Jie Song, Maxime Lefrancois, and Fabien Gandon, editors, The Semantic Web – ISWC 2019, pages 558–575, Cham, 2019. Springer International Publishing.
[23] Noémie Jacquet, Vincent Guigue, Cristina E. Manfredotti, Fatiha Saïs, Stéphane Dervaux, Paolo Viappiani: Modélisation du caractère séquentiel des repas pour améliorer la performance d'un système de recommandation alimentaire. EGC 2024: 131-142.
[24] Alexandre Combeau, Fatiha Saïs, Naggeta Kumari, Stéphane Dervaux, Cristina E. Manfredotti, Vincent Guigue, Paolo Viappiani: NutriKG - un graphe de connaissances pour modéliser les préférences et les besoins nutritionnels. IC 2025: 8-17