Health-Aware Food Recommendation Based on Knowledge Graph and Multi-Task Learning.

Area of Science:

• Artificial Intelligence
• Computer Science
• Nutrition Science

Background:

• Existing food recommender systems often fail to balance dietary preferences with personalized health requirements.
• There is a need for advanced models that can integrate complex user health data into food recommendations.

Purpose of the Study:

• To develop a novel health-aware food recommendation model (FKGM) that incorporates personalized health requirements alongside dietary preferences.
• To create a comprehensive collaborative recipe knowledge graph (CRKG) for richer food data representation.

Main Methods:

• Construction of a collaborative recipe knowledge graph (CRKG) with millions of user-recipe interactions and ingredient associations.
• Development of a score-based method to evaluate recipe healthiness against user preferences.
• Implementation of FKGM, a knowledge graph embedding and multi-task learning model utilizing a knowledge-aware attention graph convolutional neural network.

Main Results:

• FKGM demonstrated superior performance compared to four baseline models in integrating dietary preferences and personalized health requirements.
• The model achieved the best performance specifically on the health-related recommendation task.
• FKGM effectively captures semantic associations between users and recipes within the CRKG.

Conclusions:

• The proposed FKGM model offers a significant advancement in personalized healthy food recommendations.
• Integrating personalized health requirements alongside dietary preferences leads to more effective food suggestion systems.
• Knowledge graph embedding and multi-task learning are powerful techniques for building health-aware recommender systems.