JMIR Publications

ABSTRACT

Background:

Effective precision health education and promotion depend on the efficient dissemination of health information. However, current health communication encounters structural bottlenecks, including information overload with insufficient precision matching, variable quality of health resources, and a lack of personalized services. These challenges impede large-scale targeted distribution and audience access. This study aimed to develop and validate an automated tagging system using a large language model (LLM) to enhance the efficiency and equity of health communication and promotion.

Objective:

This study aimed to develop, deploy, and validate an artificial intelligence-driven, multi-tier, automated content tagging system to address the core challenges in managing Chinese health education resources and provide a technical foundation for scalable precision health communication.

Methods:

We developed a health promotion taxonomy with 10 primary, 34 secondary, and 90,562 tertiary tags using a hybrid method combining a top-down approach (aligned with national standards and expert knowledge) and a bottom-up approach (corpus mining). Subsequently, we constructed an automated tagging system for health promotion materials by fine-tuning a Baichuan2-7B LLM with Low-Rank Adaptation (LoRA), then integrated it with a named entity recognition model and a vector database (Chroma DB), and evaluated its performance.

Results:

The final taxonomy included all 16 national priority health domains. The model achieved an overall tag automation rate of 94.8% on the test set, with rates of 97.38% for text-only resources and 89.55% for nontext resources. In a comparative analysis, the model-generated tags demonstrated a higher thematic relevance to the source content than the original manual annotations.

Conclusions:

A fine-tuned LLM can efficiently automate the assignment of a granular multilevel tagging system for Chinese health promotion resources. This approach provides a scalable solution to a key bottleneck in health-information management, establishing a technical foundation for advancing precise health communication and improving equitable access to health information.