Graph Neural Networks for Health-Aware Food and Multi-Criteria Recommendation Systems

Recommender systems play a vital role in helping users discover items that align with their preferences and health needs. This thesis presents two frameworks: the Health-aware Food Recommendation System with Dual Attention in Heterogeneous Graphs (HFRS-DA) and the Multiview Graph Dual Attention and Contrastive Learning for Multi-Criteria Recommender Systems (D-MGAC). The first framework, HFRS-DA, addresses the challenge of effectively integrating heterogeneous information and uncovering meaningful relationships among entities in food and health contexts. By employing a dual hierarchical attention mechanism, HFRS-DA enhances unsupervised representation learning on heterogeneous graph-structured data. This innovative approach not only reconstructs node features and edges but also facilitates the discovery of healthy and popular recipes, promoting healthier eating habits among users. Through comprehensive analysis of both recipe components and their relationships within the heterogeneous graph, the framework identifies patterns and trends that traditional recommender systems may overlook. The second framework, D-MGAC, enhances multi-criteria recommendations by leveraging Graph Neural Networks to capture nuanced relationships between users and items. Based on a bipartite graph structure, D-MGAC treats each criterion as a separate view, utilising Dual Graph Attention combined with Contrastive Learning. This framework effectively incorporates both local (criterion-specific) and global (multi-criteria) relations. By defining anchor points in each view, we implement local and global contrastive learning to differentiate between positive and negative samples across the entire graph. Empirical evaluations on real-world datasets demonstrate that both frameworks significantly outperform existing methods, highlighting the effectiveness of attention mechanisms and contrastive learning in enhancing the accuracy of recommendations for health-aware recipes and multi-criteria items. This research advances the field of health-aware food and multi-criteria recommender systems, enabling more personalised and meaningful user experiences.

Grantor dc:publisher

UNSW, Sydney

Year dc:date

2025

Author dc:creator

• Forouzandeh, Saman

• Graph Neural Network
• Heterogeneous Information Networks
• Semantic Level Attention
• Node Level Attention
• Health Food Recommender System
• Multiview Graph Attention
• Contrastive Learning
• Multi-criteria Recommender System
• anzsrc-for: 4611 Machine learning

Statement dc:rights

• open access
• CC BY 4.0
• free_to_read

Licence

• https://purl.org/coar/access_right/c_abf2
• https://creativecommons.org/licenses/by/4.0/

Language dc:language

en

Identifier

https://doi.org/10.26190/unsworks/31317

OAI identifier oai:identifier

oai:unsworks.library.unsw.edu.au:1959.4/105379