An optimized deep learning approach for suicide detection through Arabic tweets.

Area of Science:

• Computational linguistics
• Mental health informatics
• Social media analytics

Background:

• Mental illnesses like major depressive disorder (MDD) significantly impact global well-being.
• Suicide is a leading cause of death among adolescents, highlighting the need for early detection.
• Social media platforms, including Twitter, offer a rich source of data for mental health monitoring.

Purpose of the Study:

• To develop and evaluate a framework for detecting depression and suicide risk in Arabic tweets.
• To address the gap in applying depressive detection methods to the Arabic language.
• To preprocess and categorize Arabic tweets into normal and suicide-related classes.

Main Methods:

• Scraping and annotating a dataset of Arabic tweets.
• Developing an Arabic tweet preprocessing algorithm comparing lemmatization, stemming, and lexical analysis.
• Implementing and evaluating Bidirectional Encoder Representations from Transformers (BERT) and Universal Sentence Encoder (USE) models.

Main Results:

• The proposed framework effectively categorizes Arabic tweets for mental health analysis.
• Arabic BERT models achieved a best weighted sum metric (WSM) of 95.26%.
• Universal Sentence Encoder (USE) models achieved a best WSM of 80.2%.

Conclusions:

• The study demonstrates the feasibility of using NLP models for mental health monitoring in Arabic social media.
• The developed framework and preprocessing techniques show promise for early detection of depression and suicide risk.
• Further research can build upon these findings to enhance mental health support through digital platforms.