Boamente: A Natural Language Processing-Based Digital Phenotyping Tool for Smart Monitoring of Suicidal Ideation

Area of Science:

• Digital Mental Health
• Computational Linguistics
• Machine Learning in Healthcare

Background:

• Individuals at risk of suicide often experience isolation, hindering traditional monitoring methods.
• Timely identification of suicidal ideation is crucial for effective mental health interventions.
• Existing methods for monitoring suicidal ideation face challenges due to the private nature of users' thoughts.

Purpose of the Study:

• To develop and validate the Boamente tool for passively identifying suicidal ideation from smartphone text data.
• To enable early detection of suicidal ideation for prompt intervention by mental health professionals.
• To leverage natural language processing and deep learning for enhanced mental health monitoring.

Main Methods:

• A virtual keyboard mobile application passively collected user text data.
• Text data was processed on a web platform using natural language processing and deep learning models.
• Various machine learning and deep learning algorithms were evaluated for text classification accuracy.
• The BERTimbau Large model was selected based on performance metrics in a validation study.

Main Results:

• The Boamente tool successfully identified suicidal ideation from user texts.
• The BERTimbau Large model achieved high performance, with a recall of 0.953.
• Key performance metrics included accuracy (0.955), precision (0.961), F-score (0.954), and AUC (0.954).
• The tool's effectiveness was demonstrated in studies involving mental health professionals and patients.

Conclusions:

• The Boamente tool offers a novel approach to monitoring suicidal ideation through passive text data collection.
• The developed deep learning model shows significant potential for real-time identification of at-risk individuals.
• This technology can support mental health professionals by providing timely insights for patient intervention.