Beyond Similarity: Heterogeneous Graph Learning for Multi-Objective Food Substitution in Charitable Food Agencies

Naimur Rahman Chowdhury; Limon Bin Hossain

cs.AIAug 22, 2026

Beyond Similarity: Heterogeneous Graph Learning for Multi-Objective Food Substitution in Charitable Food Agencies

Authors: Naimur Rahman Chowdhury, Limon Bin Hossain

Organizations: Department of Industrial and Systems Engineering North Carolina State University Raleigh, NC, USA · Department of Industrial and Production Engineering Bangladesh University of Engineering and Technology Dhaka, Bangladesh

Abstract

Charitable food agencies play an important role in alleviating food insecurity by distributing donated food to people in need. However, they rely on ad hoc in-kind donations and often face shortages of specific foods, so they offer substitutes. A good food substitution requires matching household preferences, nutritional needs, and item similarity. Agencies have limited direct records of consumption behavior due to resource constraints, making it challenging to make an appropriate substitution decision that meets multiple criteria. In this study, we propose a heterogeneous graph neural network (HeteroGNN), a source-grounded recommendation framework for food substitution in charitable food agencies. We first build a unified relational graph from large-scale public data sources, combining household behavior on food consumption and food nutrient information in the United States (US) context. We treat the substitution recommendation as a multi-objective ranking problem with three targets, including behavior affinity, health suitability, and substitution similarity. We train and validate the proposed framework under standard graph relationship and adverse cold-start settings by removing relational edges from the graph. Our results show that the proposed framework leverages relational information beyond node features in predicting consumption behavior. Additionally, the proposed framework remains robust with sparsity when the model receives incomplete information about behavior and nutrient features. Finally, we show the weak correlation among different objectives, thereby justifying the multi-objective framing as a replacement for an aggregated decision. The proposed framework can help downstream charitable agency decision-makers make contextspecific substitution recommendations with limited information available.

Figures & tables

Fig. 1 : Complete data construction with food catalog, population and behavior relations for heterogeneous graph.

Quantity	Count
People	26,250
Household/respondent contexts	16,759
Demographic household groups	42
Foods with nutrient evidence	9,903
Food–nutrient records	102,441
Consumption lines	188,148

TABLE I : Scale of the data in the graph construction.

Objective	Supervision	Interpretation
Behavior affinity	Observed group–food relation vs. sampled unobserved relation	The probability that the food is familiar or behaviorally plausible for the household group.
Health suitability	Profile-weighted food-health score	Nutrient-density suitability for the health-profile mix associated with the group.
Substitution similarity	Directional mapped catalog-pair score	The degree to which a candidate can play the same role as the requested item.
Composite ranking	Weighted combination over the three predictions	Inspectable decision support rather than a single uninterpretable score.

TABLE II : Interpretation of the multi-objective recommendation.

Node type	Feature name	Description	Encoding type
Household group	Member count	Number of respondent contexts in the group	Discrete
Household group	Mean household size	Average household size within the group	Discrete
Household group	Mean income-to-poverty ratio	Average income-to-poverty ratio within the group	Continuous
Household group	SNAP share	Share of group members participating in SNAP	Continuous proportion
Household group	Poverty band	Discretized income-to-poverty category	One-hot binary
Household group	Household-size band	Discretized household-size category	One-hot binary

TABLE III : Node feature construction for the heterogeneous graph.

Hyperparameter	Value
Node embedding dimension	64
Message-passing layers	2
Activation function	ReLU
Aggregation function	Sum
Dropout	0.218
Optimizer	AdamW

TABLE IV : Model hyperparameters and training Settings.

Metric		HeteroGNN	MLP	BPR	LightGCN
Behavior affinity	ROC-AUC	0.9480	0.9009	0.9144	0.9369
	PR-AUC	0.9027	0.7809	0.8467	0.8838
	Precision@5	0.9952	0.8000	0.9667	0.9667
	Recall@5	0.0262	0.0186	0.0244	0.0240
	NDCG@5	0.9965	0.8039	0.9724	0.9699
Health suitability	MAE	0.0182	0.0128	0.0756	0.0765

TABLE V : Performance comparison under the standard evaluation setting. Best results in bold .

Metric		HeteroGNN		MLP		BPR		LightGCN
		CS-F	CS-G	CS-F	CS-G	CS-F	CS-G	CS-F	CS-G
Behavior affinity	ROC-AUC	0.8381	0.6070	0.8973	0.8971	0.4959	0.3883	0.4887	0.5323
	PR-AUC	0.6932	0.4367	0.7739	0.7618	0.3347	0.2749	0.3511	0.3910
	Precision@5	0.5143	0.2500	0.7857	0.8750	0.3333	0.1500	0.2238	0.3750
	Recall@5	0.0108	0.0008	0.0179	0.0024	0.0057	0.0006	0.0039	0.0009
	NDCG@5	0.4656	0.2500	0.8012	0.8453	0.3390	0.1517	0.2405	0.3750

TABLE VI : Performance comparison under the cold-start evaluation settings. CS-F = cold-start food; CS-G = cold-start household group.

Attribute	Level	ROC-AUC
Poverty band	Above 2 $\times$ poverty	0.470
	Below poverty	0.781
	1–2 $\times$ poverty	0.667
SNAP band	Non-SNAP	0.590
SNAP band	SNAP	0.619
Household size	Large	0.640

TABLE VII: Behavior Affinity Performance by Household Demographic Segment (Cold-Start Group Split)

	Behavior Sparsity				Nutrient Sparsity
	Behavior ROC AUC $\uparrow$		Health MAE $\downarrow$		Behavior ROC AUC $\uparrow$		Health MAE $\downarrow$
Removed	GNN	MLP	GNN	MLP	GNN	MLP	GNN	MLP
0%	0.9444	0.8890	0.0285	0.0262	0.9444	0.8890	0.0285	0.0262
10%	0.9474	0.8889	0.0283	0.0263	0.9433	0.8895	0.0329	0.0254
30%	0.9442	0.8889	0.0290	0.0265	0.9388	0.8890	0.0327	0.0285
50%	0.9378	0.8889	0.0292	0.0265	0.9469	0.8876	0.0379	0.0481

TABLE VIII : Robustness of HeteroGNN and MLP in incomplete information.

	Pearson			Spearman
	Behavior	Health	Similarity	Behavior	Health	Similarity
Behavior affinity	1.000	$-$ 0.067	0.090	1.000	0.521	0.311
Health suitability	$-$ 0.067	1.000	0.070	0.521	1.000	0.421
Substitution similarity	0.090	0.070	1.000	0.311	0.421	1.000

TABLE IX : Correlations among the three objectives.

Recommender Systems

Recommender systems predict and rank items for people using interaction histories, content, context, and feedback.

See all 516 papers

Jun 22, 2026

MORL-A2C: Multi-Objective Reinforcement Learning Reranker for Optimizing Healthiness in MOPI-HFRS

Unhealthy dietary behavior continues to be a persistent public health issue in the United States, exacerbated by recommendation systems that prioritize user preference without considering nutritional health. The Multi-Objective Personalized Interpretable Health-aware Food Recommendation System (MOPI-HFRS), from which this work extends, addresses this by jointly optimizing preference, health, and diversity through Pareto-based optimization. However, this approach relies on static, per-step tradeoff solutions that fail to capture the sequential nature of dietary decision-making. We introduce MORL-A2C, a sequential decision-making extension to MOPI-HFRS targeting the health-preference axis. Leveraging frozen GNN embeddings, MORL-A2C formulates recommendation as a K-step reranking problem using an Advantage Actor-Critic algorithm with a scalarized relevance/health reward. The policy is warm-started via behavior cloning against a dot-product ranker derived from frozen embeddings. We also identify and correct a non-trivial bug in the MOPI-HFRS evaluation pipeline that understated baseline performance; all results are reported against the corrected baseline. On the macro-nutrient benchmark, MORL-A2C achieves a modest reduction in ranking quality (Recall@20: 25.64% to 23.61%, NDCG@20: 23.52% to 20.64%) in exchange for a substantial improvement in health alignment (H-Score@20: 46.05% to 69.57%), with consistent trends on the full-nutrient benchmark. These findings validate that policy-driven sequential optimization can effectively navigate the health-preference trade-off in multi-objective food recommendation.

Aarya Vasantlal, Joshua Zolla, Chuxu Zhang

University of Connecticut, Storrs, CT, USA

Multi-Objective Reinforcement Learning Sequential Decision Making Recommender Systems

Ask PDF

Oct 22, 2026

FairFRESH: A Neuro-Symbolic Framework for Fair and Explainable Household Food Recommendation

Recommender systems typically serve one user at a time. However, in households, members share a finite pantry and follow different dietary rules, making independent recommendations prone to resource conflicts and dietary violations. We formulate household food recommendation as constrained fair division and present FairFRESH, a neuro-symbolic framework that learns per-member utilities, allocates foods under explicit welfare objectives, enforces dietary rules through a deterministic filter, and explains recommendations using Shapley and Banzhaf attributions. Theoretically, we show that no finite penalty guarantees compliance near a dietary-rule threshold, establish approximation guarantees for greedy allocation under utilitarian and smoothed Nash objectives in the stock-free model before filtering, and characterize the scale invariance of Nash welfare. We also use the minimax theorem to analyze rotating shared meals. On NHANES-derived data, the learned utility model achieves Recall@20 of 0.293, outperforming five learned recommenders (at most 0.244) but not a previous-day repetition heuristic (0.321). On held-out four-person households, Nash allocation reduces Gini disparity by 42% relative to conflict resolution while improving Recall@20 from 0.180 to 0.197. It comes within 2% of an upper bound on the smoothed Nash optimum, and rescaling one member's utilities changes at most 1.5% of assignments. All recommendations satisfy the encoded dietary rules, while attribution scores correlate with utility loss upon component removal (r = 0.76). In simulation, shared meals reduce pantry use by about one-third, and weekly rotation increases the worst-off member's utility by 38% over the best fixed meal.

Kodur Sai Vinay Sathvik, Kakoli Bora, Tejaswini R. M

Department of Artificial Intelligence and Machine Learning, BNM Institute of Technology

Fair Resource Allocation Social Welfare Maximization Explainable Recommendation Neuro-Symbolic AI

Ask PDF

Sep 9, 2024

Revealing the Pitfalls and Re-Evaluating the Advancement of Heterophilic Graph Learning

Over the past decade, Graph Neural Networks (GNNs) have achieved great success on machine learning tasks with relational data. However, recent studies have found that heterophily can cause significant performance degradation of GNNs, especially on node-level tasks. Numerous heterophilic benchmark datasets have been put forward to validate the efficacy of heterophily-specific GNNs, and various homophily metrics have been designed to help recognize these challenging datasets. Nevertheless, there still exist multiple pitfalls that severely hinder the proper evaluation of new models and metrics: 1) lack of hyperparameter tuning; 2) insufficient evaluation on the truly challenging heterophilic datasets; 3) missing quantitative evaluation for homophily metrics on synthetic graphs. To overcome these challenges, we first train and fine-tune baseline models on

27

most widely used benchmark datasets, and categorize them into three distinct groups: malignant, benign and ambiguous heterophilic datasets. We identify malignant and ambiguous heterophily as the truly challenging subsets of tasks, and to our best knowledge, we are the first to propose such taxonomy. Then, we re-evaluate

11

state-of-the-arts (SOTA) GNNs, covering six popular methods, with fine-tuned hyperparameters on different groups of heterophilic datasets. Based on the model performance, we comprehensively reassess the effectiveness of different methods on heterophily. At last, we evaluate

11

popular homophily metrics on synthetic graphs with three different graph generation approaches. To overcome the unreliability of observation-based comparison and evaluation, we conduct the first quantitative evaluation and provide detailed analysis.

Sitao Luan, Qincheng Lu, Chenqing Hua +3

Mila - Quebec Artiﬁcial Intelligence Institute · 1McGill University

Graph Neural Networks Node Classification Graph Heterophily

Ask PDF

Abstract

Recommender Systems

Explore similar work

MORL-A2C: Multi-Objective Reinforcement Learning Reranker for Optimizing Healthiness in MOPI-HFRS

FairFRESH: A Neuro-Symbolic Framework for Fair and Explainable Household Food Recommendation

Revealing the Pitfalls and Re-Evaluating the Advancement of Heterophilic Graph Learning