JMIR Publications

ABSTRACT

Background:

Nutrient needs vary over the lifespan. Improving knowledge of both population groups and care providers can help make healthier food choices, thereby promoting population health and preventing diseases. Providing evidence-based food knowledge online is a credible, low cost, and easily accessible way.

Objective:

This study aims to develop an online multimodal food data exploration platform for easily access evidence-based diet- and nutrition-related data.

Methods:

We developed an online food data exploration platform, Food Atlas, which was constructed based on multimodal Chinese food knowledge graph. Firstly, we chose pregnant women as our target population, and manually curated the knowledge graph from evidence-based resources. Then, various interactions with graph-structured data for easy access were developed, including graph-based interactive visualization, natural language interface, and multimodal image-text retrieval.

Results:

The constructed multimodal food knowledge graph totally contained 2011 entities, 10410 triplets and 23497 images. Its schema consisted of eleven entity types and twenty-six types of semantic relations. The results showed that Food Atlas could support diversified food information retrieval for different user groups, including pregnant women, clinicians, and dietitians. A clinician could use our natural language interface to input the query “哪些食物可以减轻孕吐？” (What foods can relieve morning sickness?), and found that “蜂蜜柚子茶” (honey citron tea) was the food he was looking for.

Conclusions:

Food Atlas we developed can optimize the organization, presentation, and interaction of existing evidence-based multimodal dietary data. It has the potential feasibility to help populations at different life stage make healthier food choices.