Automatic detection of depression symptoms in twitter using multimodal analysis

Area of Science:

• Computational social science
• Mental health informatics
• Natural language processing

Background:

• Depression is a leading cause of suicide, and social media offers insights into psychological states.
• Identifying mental health conditions early is crucial for intervention and support.

Purpose of the Study:

• To present an automated, multimodal framework for predicting depression symptoms using social media user profiles.
• To evaluate the effectiveness of various computational methods in detecting depression from social data.

Main Methods:

• Utilized n-gram language models, LIWC dictionaries, automatic image tagging, and bag-of-visual-words for feature extraction.
• Employed correlation-based feature selection and nine different classifiers for model evaluation.
• Collected and analyzed tweets and user bio-text for self-reported statements indicative of depression.

Main Results:

• Tweets alone achieved 91% accuracy in predicting depressive symptoms.
• User bio-text alone demonstrated 83% accuracy in identifying depression symptoms.
• The multimodal framework showed promising results in assessing psychological states from social media.

Conclusions:

• Social media data, particularly tweets and bios, can be effectively used for automated depression symptom detection.
• Further performance improvements are anticipated with domain-specific user data or clinical information integration.
• This approach offers a novel pathway for large-scale mental health monitoring and early detection.