AWARE Narrator and the Utilization of Large Language Models to Extract Behavioral Insights from Smartphone Sensing Data
This work addresses the need for more interpretable and comprehensive behavioral analysis in digital health, particularly for mental health assessment, though it appears incremental as it builds on existing digital phenotyping methods by adding narrative generation.
The paper tackles the problem of extracting behavioral insights from smartphone sensing data by introducing AWARE Narrator, a framework that converts quantitative sensor data into structured English narratives, and demonstrates its application on a week of data from university students to summarize behavior and analyze psychological states using large language models.
Smartphones, equipped with an array of sensors, have become valuable tools for personal sensing. Particularly in digital health, smartphones facilitate the tracking of health-related behaviors and contexts, contributing significantly to digital phenotyping, a process where data from digital interactions is analyzed to infer behaviors and assess mental health. Traditional methods process raw sensor data into information features for statistical and machine learning analyses. In this paper, we introduce a novel approach that systematically converts smartphone-collected data into structured, chronological narratives. The AWARE Narrator translates quantitative smartphone sensing data into English language descriptions, forming comprehensive narratives of an individual's activities. We apply the framework to the data collected from university students over a week, demonstrating the potential of utilizing the narratives to summarize individual behavior, and analyzing psychological states by leveraging large language models.