Explainable Predictive Model for Suicidal Ideation During COVID-19: Social Media Discourse Study.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Mental Health Research

Background:

• COVID-19 pandemic significantly impacted mental health, necessitating new methods for monitoring.
• Traditional mental health assessments face challenges with stigma and accessibility.
• Social media offers a platform to analyze language patterns related to suicidal ideation.

Purpose of the Study:

• To detect suicidal ideation by analyzing social media text using Natural Language Processing (NLP).
• To identify evolving language patterns and contributing factors to suicidal ideation during the pandemic.

Main Methods:

• A hybrid deep learning model combining Bidirectional Encoder Representations from Transformers (BERT), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) was developed.
• Features were extracted using Term Frequency-Inverse Document Frequency (TF-IDF), Word2vec, and BERT.
• Explainable Artificial Intelligence (XAI) techniques, including LIME and SHAP, were employed to identify key contributing factors.

Main Results:

• The hybrid BERT+CNN+LSTM model achieved high performance: 94% precision, 95% recall, 94% F1-score, and 93.65% accuracy.
• Analysis revealed a shift in language features associated with suicidal ideation during and after the COVID-19 pandemic.

Conclusions:

• The proposed fused architecture effectively captures contextual information for understanding and predicting suicidal ideation.
• The COVID-19 pandemic introduced new linguistic features linked to increased suicidal tendencies.
• Developing targeted strategies is crucial to address the rise in mental health challenges exacerbated by the pandemic.