ContextVecNet: A Context-Driven Multimodal Learning Framework for Depression Detection.

Area of Science:

• Computational linguistics
• Artificial intelligence
• Mental health informatics

Background:

• Depression is a significant global health concern.
• Social media data offers potential for early depression detection.
• Existing multimodal approaches lack effective context and temporal analysis.

Purpose of the Study:

• To propose ContextVecNet, a novel multimodal deep learning framework.
• To enhance the accuracy and reliability of depression detection from social media.
• To address limitations in capturing contextual relationships and temporal information.

Main Methods:

• Developed ContextVecNet, a CLIP-based architecture with learnable context vectors.
• Integrated context vectors into text and image encoding.
• Incorporated a cross-modal transformer with time-aware embeddings for temporal dynamics and cross-modal interactions.

Main Results:

• ContextVecNet achieved state-of-the-art performance on a multimodal Twitter dataset.
• Achieved an Area Under the Curve (AUC) of 0.9922 and an F1-score of 0.9619.
• Ablation study confirmed the critical role of context vectors in performance.

Conclusions:

• ContextVecNet effectively models temporal dynamics and cross-modal interactions for depression detection.
• The framework demonstrates superior performance compared to existing methods.
• Learnable context vectors are crucial for adapting to specific depression markers in social media data.