Detecting Depression with Smartphones... KAIST "Early Diagnosis Without Privacy Invasion"

A technology capable of detecting depression through smartphones has been developed. It measures depressive feelings by analyzing users' language patterns, with the advantage of enabling early diagnosis without invading privacy.

A schematic diagram of technology that diagnoses mental health status based on voice and keyboard input from smartphone users. Provided by KAIST

On the 21st, KAIST announced that a research team led by Professor Seongju Lee from the Department of Electrical Engineering developed an artificial intelligence technology that can analyze and diagnose mental health conditions simply by users carrying and routinely using their smartphones.

The research team began their study based on the fact that mental illness diagnosis is typically conducted by analyzing language use during consultations with patients.

The developed technology focuses on diagnosing mental health by reflecting these characteristics, using keyboard input such as text messages written directly by users and real-time voice data collected from the smartphone's microphone.

Until now, language data contained sensitive user information, limiting its use. Concerns over privacy invasion hindered the utilization of language data.

To address this, the research team applied federated learning AI technology, enabling the AI model to be trained and perform diagnosis without leaking data from individual users' devices, thereby resolving privacy concerns.

Professor Seongju Lee (first from the left) is taking a commemorative photo with master's and doctoral students from the research team. Photo by KAIST

The AI model developed by the research team is characterized by being trained on datasets accumulated based on everyday conversation content and the speaker's mental health. It analyzes conversations input and exchanged on smartphones in real time and predicts the user's mental health scale based on the learned information.

Furthermore, the research team developed a methodology to perform mental health diagnosis based on user language data.

Recognizing that users' language patterns vary depending on the situation, the core design enables the AI model to focus on relatively important language data based on current situational cues.

For example, if the AI determines that more diagnostic clues can be found during evening hours or conversations with family or friends rather than during work hours, it concentrates and analyzes monitoring under the same conditions.

The research team's achievements (paper) were also presented at EMNLP (Conference on Empirical Methods in Natural Language Processing), held in Singapore from the 6th to the 10th of this month. EMNLP is regarded as the most prestigious conference in the field of natural language processing.

Professor Seongju Lee said, "We expect our research results to provide opportunities for those struggling with depression and similar issues to diagnose and respond to their mental health without concerns about personal information leakage or privacy invasion. We hope that this research outcome will be commercialized and provide practical benefits to society."

This content was produced with the assistance of AI translation services.