Depression detection for twitter users using sentiment analysis in English and Arabic tweets

Area of Science:

• Computational linguistics
• Mental health informatics
• Natural language processing

Background:

• Depression poses significant risks, including suicidal ideation and daily disability.
• Early detection of depression is crucial for timely intervention and reducing mortality.
• Social media platforms offer a rich source for identifying early signs of depression.

Purpose of the Study:

• To develop and evaluate machine learning models for detecting depression in Arabic and English social media text.
• To create annotated corpora for training and validating depression detection models.
• To build a web application for real-time depression detection and trend prediction.

Main Methods:

• Development of five machine learning models for depression detection using Twitter data.
• Creation of a manually annotated Arabic corpus (Arabic_Dep_tweets_10,000).
• Generation of two automatically annotated English corpora (Eng_without_negation_60.000 and Eng_with_negation_57.000).

Main Results:

• The best Arabic model achieved a 96.6% f1-score for binary classification.
• For English text without negation, models reached 92% (binary) and 88% (multi-class) f1-scores.
• For English text with negation, models achieved 87% (binary) and 85% (multi-class) f1-scores.

Conclusions:

• Machine learning models can effectively detect depression from social media text in both Arabic and English.
• The developed corpora and models facilitate large-scale depression screening and analysis.
• A web application is available for practical implementation of depression detection using social media data.