All topics | KAIST Breakthroughts

KAIST
BREAKTHROUGHS

Research Webzine of the KAIST College of Engineering since 2014

Spring 2024 Vol. 22

Electronics

Privacy-preserving mental health monitoring with smartphones

February 27, 2024 hit 622

Fedtherapist, a mental health monitoring system with user-generated linguistic expressions on smartphones via federated learning

A research team at KAIST, led by Professor Sung-Ju Lee, proposed FedTherapist, a privacy-preserving smartphone mental health monitoring system that uses federated learning for continuous speech and keyboard input.

Figure 1. An overview of FedTherapist, a mental health monitoring system with user-generated linguistic expressions on smartphones leveraging federated learning

The research team led by Professor Sung-Ju Lee in the School of Electrical Engineering developed a technology named FedTherapist that automatically analyzes the language usage patterns of users on smartphones to monitor their mental health status. This system enables analysis of the users' mental health status simply through their everyday smartphone use.

FedTherapist was designed based on the observation that psychiatrists diagnose mental disorders by conversing with patients. FedTherapist works by analyzing (1) the content of keyboard inputs such as SMS messages written by the user and (2) the speech of a user captured from the voice data collected from the smartphone's microphone.

Previously, using such language data had been challenging due to substantial privacy concerns. To address this issue, the research team adopted federated learning, which collaboratively trains the AI model on the user’s device without data collection, avoiding privacy concerns.

Furthermore, the team has developed a methodology to perform effective mental health monitoring from the massive amount of user language data available on smartphones. The AI model is designed to focus on the user-generated text in a context in which users are highly likely to expose their mental health status. For example, users would be more likely to express their emotions via text at night at home than daytime at the workplace. Such a context (time, location, and user activity.) is based on the information captured on the smartphone and its embedded sensors.

This research is a joint effort involving Jaemin Shin (KAIST Ph.D. candidate), Hyungjun Yoon (KAIST Ph.D. candidate), Seungjoo Lee (KAIST MS candidate), Professor Sung-Ju Lee, Dr. Sungjun Park (CEO of SoftlyAI/KAIST Ph.D.), Professor Yunxin Liu (Tsinghua University), and Professor Jinho D. Choi (Emory University).

This technology was presented at Conference on Empirical Methods in Natural Language Processing (EMNLP) 2023, a leading conference in the field of natural language processing.

※ Paper Title: FedTherapist: Mental Health Monitoring with User-Generated Linguistic Expressions on Smartphones via Federated Learning

Paper link: https://aclanthology.org/2023.emnlp-main.734.pdf

LIST