MOPI-HFRS: A Multi-objective Personalized Health-aware Food Recommendation System with LLM-enhanced Interpretation Conference: KDD ’25: SIGKDD Conference on Knowledge Discovery and Data Mining; ;

Zheyuan Zhang, Zehong Wang, Tianyi Ma, Varun Sameer Taneja, Sofia Nelson, Nhi Ha Lan Le, Keerthiram Murugesan, Mingxuan Ju, Nitesh V. Chawla, Chuxu Zhang, Yanfang Ye · 2024

Read MOPI-HFRS: A Multi-objective Personalized Health-aware Food Recommendation System with LLM-enhanced Interpretation Conference: KDD ’25: SIGKDD Conference on Knowledge Discovery and Data Mining; ; by Zheyuan Zhang, Zehong Wang, Tianyi Ma, Varun Sameer Taneja, Sofia Nelson, Nhi Ha Lan Le, Keerthiram Murugesan, Mingxuan Ju, Nitesh V. Chawla, Chuxu Zhang, Yanfang Ye (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-08847.txt.

Introduction

Zheyuan Zhang and co-authors build a health aware food recommendation system that explains itself, in a 2024 paper. The catalog title adds conference metadata, so the head title governs. Big platforms rank food by taste alone, though poor eating habits worry the United States. Past health aware systems rarely tuned advice to a person's own conditions, and few explained their picks.

The team first built two large benchmarks for personal, health aware food advice. Their system then weighs three goals at once: taste, personal health fit, and nutrient variety. A large language model adds plain reasons for each pick, drawn from healthy eating knowledge. The aim is advice people can judge and trust, not just obey.

Catalog record

CREATORZheyuan Zhang, Zehong Wang, Tianyi Ma, Varun Sameer Taneja, Sofia Nelson, Nhi Ha Lan Le, Keerthiram Murugesan, Mingxuan Ju, Nitesh V. Chawla, Chuxu Zhang, Yanfang Ye
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-08847

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

MOPI-HFRS: A Multi-objective Personalized Health-aware Food Recommendation System with LLM-enhanced Interpretation Conference: KDD ’25: SIGKDD Conference on Knowledge Discovery and Data Mining; ; Zheyuan Zhang, Zehong Wang, Tianyi Ma, Varun Sameer Taneja, Sofia Nelson, Nhi Ha Lan Le, Keerthiram Murugesan, Mingxuan Ju, Nitesh V. Chawla, Chuxu Zhang, Yanfang Ye Abstract . The prevalence of unhealthy eating habits has become an increasingly concerning issue in the United States. However, major food recommendation platforms (e.g., Yelp) continue to prioritize users’ dietary preferences over the healthiness of their choices. Although efforts have been made to develop health-aware food recommendation systems, the personalization of such systems based on users’ specific health conditions remains under-explored. In addition, few research focus on the interpretability of these systems, which hinders users from assessing the reliability of recommendations and impedes the practical deployment of these systems. In response to this gap, we first establish two large-scale personalized health-aware food recommendation benchmarks at the first attempt . We then develop a novel framework, M ulti- O bjective P ersonalized I nterpretable H ealth-aware F ood R ecommendation S ystem ( MOPI-HFRS ), which provides food recommendations by jointly optimizing the three objectives: user preference, personalized healthiness and nutritional diversity, along with an large language model (LLM)-enhanced reasoning module to promote healthy dietary knowledge through the interpretation of recommended results. Specifically, this holistic graph learning framework first utilizes two structure learning and a structure pooling modules to leverage both descriptive features and health data. Then it employs Pareto optimization to achieve designed multi-facet objectives. Finally, to further promote the healthy dietary knowledge and awareness, we exploit an LLM by utilizing knowledge-infusion, prompting the LLMs with knowledge obtained from the recommendation model for interpretation. Extensive experimental results based on our established benchmarks demonstrate the outstanding performance of MOPI-HFRS in providing both diverse, healthy food recommendations and reliable explanations by comparison with state-of-the-art baseline methods. Our code, built benchmarks are available at here . Figure 1. The illustration of the impact of a poor diet, and the comprehensive and challenging nature of the multi-objective personalized interpretable health-aware food recommendation task. 1. Introduction Diet plays a pivotal role in maintaining and enhancing human health. A balanced diet provides the essential nutrients required for optimal physical and mental well-being, while poor dietary habits can lead to numerous adverse health outcomes, highlighting the critical importance of diet in public health ( Organization 2021 ) . Despite the well-documented benefits of a balanced healthy diet, the prevalence of unhealthy eating habits remains a significant concern in modern society. Approximately 42.4% of adults in the United States are classified as obese ( for Disease Control and Prevention 2020 ) , and in 2017 alone, 11 million deaths and disability-adjusted life-years (DALYs) were attributable to dietary risk factors, such as high intake of sodium ( Afshin et al. 2019 ; Organization 2023 ) . It is vital to counteract this trend and encourage individuals to adopt healthier eating habits. Unfortunately, the field of health-aware food recommendation systems has not received as much attention as other domains, such as movie recommendation systems ( Trattner and Elsweiler 2017 ) . This disparity may be due to the comprehensive and challenging nature of the task: On one hand, there is not a one-size-fits-all solution for healthy food recommendations, and the system needs to balance multiple goals, such as user preference, personalized healthiness, and diversity, as depicted in Figure- 1 (a); On the other hand, the explainability of the system is crucial, as the goal of such systems is not only to inform users of the recommendations but also to explain why the options are healthier, thereby raising their awareness and knowledge towards healthy diets, as shown in Figure- 1 (b). So the question naturally occurs: How can we develop an effective and interpretable food recommendation system that simultaneously considers users’ dietary preferences, diversity, and personal health conditions? To answer this question, numerous existing studies ( Ge et al. 2015 ; Shirai et al. 2021 ; Wang et al. 2021 ; Bölz et al. 2023 ; Zhang et al. 2024 ) have attempted to define and measure the healthiness of foods. However, we argue these efforts are significantly limited in several ways, which are unable to fully address the complexities of our question at hand: First, to the best of our knowledge, none of the current works are truly personalized to users’ specific health conditions, because of the inaccessibility of users’ medical information. The definition of "healthy" varies significantly between individuals with different health profiles, such as the individuals with high body mass index (BMI) compared to those with low BMI, or individuals recovering from opioid misuse compared to those with chronic kidney diseases. Second, the healthiness of a food in current studies tend to be used as a hard-coded post-processing method, with the information rarely utilized during the training process, thus unable to capture more nuanced interplay between health factors and user preference. Third, existing approaches often neglect the importance of balancing health goals with individual tastes, which hinders the practical deployment of such systems, as this can lead to potential user dissatisfaction. Finally, relatively few studies focus on the interpretability of the models, whereas the interpretations are crucial in health eating habit promotion, as they help user understand and accept the recommendations, and empower the users to make better-informed decisions about their diet beyond the immediate use of the recommendation system in the future. To overcome the aforementioned four research gaps, we first establish two large-scale food recommendation benchmark datasets, marking the first attempt to integrate users’ medical data for personalized health-aware food recommendations. Specifically, we leverage data from the National Health and Nutrition Examination Survey (NHANES) ( CDC 2024 ) , a publicly available dataset designed to evaluate the health and nutritional status of adults and children in the United States. Using this data, we construct the Health and Nutrition Recommendation Bipartite Graph and two benchmark datasets derived from this graph to comprehensively integrate users’ health-related information and food records. This way we tackle the first research gap of unable to personalize the recommendation. We then develop a novel framework, M ulti- O bjective P ersonalized I nterpretable H ealth-aware F ood R ecommendation S ystem ( MOPI-HFRS ), that bridges the gap between graph-based recommendation system and LLMs to train an interpretable food recommendation system that simultaneously considers three objectives: user preference, personalized healthiness, and nutritional diversity. Specifically, MOPI-HFRS is designed to address the remaining three gaps as follows: First, to tackle the issue of not involving health information in the training, we develop a health-aware graph structural learning module that dynamically incorporates both health-related and descriptive feature information during training. Second, to achieve the optimal recommendations while balancing the three objectives, we apply Pareto multi-objective learning for model optimization. Finally, to increase system’s interpretability, we further integrate the generative power of LLMs that prompted with knowledge obtained from the recommendation tasks for explanations, raising public awareness and knowledge about healthy diets. Overall, the key contributions of this work are summarized as follows: • New Benchmarks. In this work, we construct a large-scale Health and Nutrition Recommendation Graph and establish two benchmark datasets from the graph for health-aware

Read the complete text