Paper Type

Short

Paper Number

PACIS2025-1150

Description

In recent years, depression has emerged as one of the leading causes of suicide. The widespread use of social media platforms and the anonymous characteristic of online communication provide individuals with depression a channel to express their emotions. This also presents a valuable opportunity to detect potential cases of depression promptly. However, current machine learning-based depression detection methods still suffer from limited performance and reliance on single-modal data. To address these limitations, we propose a multi-modal depression detection model (MDD) that leverages the multi-head attention mechanism to process features from text, images, and time data generated by users. The experimental results show that MDD has excellent performance in depression detection, achieving F1 score of 91.38%, precision of 90.86%, and recall of 91.90%. Our approach provides an effective method of user-level depression detection by considering multiple forms of user-generated content in social media and can better identify depressed patients.

Comments

Healthcare

Share

COinS
 
Jul 6th, 12:00 AM

A Multimodal Approach to Depression Detection from Social Media Data

In recent years, depression has emerged as one of the leading causes of suicide. The widespread use of social media platforms and the anonymous characteristic of online communication provide individuals with depression a channel to express their emotions. This also presents a valuable opportunity to detect potential cases of depression promptly. However, current machine learning-based depression detection methods still suffer from limited performance and reliance on single-modal data. To address these limitations, we propose a multi-modal depression detection model (MDD) that leverages the multi-head attention mechanism to process features from text, images, and time data generated by users. The experimental results show that MDD has excellent performance in depression detection, achieving F1 score of 91.38%, precision of 90.86%, and recall of 91.90%. Our approach provides an effective method of user-level depression detection by considering multiple forms of user-generated content in social media and can better identify depressed patients.