Neural Models for Generating Natural Language Summaries from Temporal Personal Health Data.

Area of Science:

• Digital Health
• Artificial Intelligence
• Behavioral Science

Background:

• Growing interest in personal health technologies generates vast amounts of user data.
• Existing methods for summarizing temporal health data often rely on rule-based approaches.
• Bridging the gap between raw data and actionable user insights is crucial for health management.

Purpose of the Study:

• To explore the use of advanced machine learning models for automatic natural language summary generation from personal health data.
• To assess the effectiveness of recurrent, convolutional, and Transformer-based encoder-decoder architectures for this task.
• To enable users to gain meaningful behavioral insights from their tracked health information.

Main Methods:

• Utilized recurrent, convolutional, and Transformer-based encoder-decoder models.
• Trained and evaluated models on real-world temporal personal health data.
• Focused on generating natural language summaries from numeric time-series data.

Main Results:

• Demonstrated the capability of the developed models to automatically generate high-quality natural language summaries.
• Showcased the effectiveness of the models using data from MyFitnessPal.
• Achieved successful summarization of complex temporal personal health data.

Conclusions:

• The study presents a significant advancement in automatically generating temporal summaries from personal health data.
• The developed models offer a promising approach to enhance user understanding of their health behaviors.
• This work lays the foundation for future research in creating novel and meaningful insights from personal health informatics.