Depression Detection Based on Hybrid Deep Learning SSCL Framework Using Self-Attention Mechanism: An Application to...

Area of Science:

• Computational linguistics
• Artificial intelligence
• Mental health informatics

Background:

• Depression is a significant global health issue, impacting millions worldwide.
• Social media platforms offer a rich source of user-generated text data for mental health analysis.
• Existing methods for depression detection from text often struggle with implicit or nuanced expressions.

Purpose of the Study:

• To develop and validate a novel deep learning framework for accurate depression detection from textual data.
• To capture both explicit and implicit indicators of depression in social media posts.
• To evaluate the framework's performance across different datasets and tasks, including cross-domain validation.

Main Methods:

• A hybrid deep learning model named Sequence, Semantic, Context Learning (SSCL) was developed.
• The framework utilizes GloVe for word embeddings, LSTM and CNN for sequence and semantic understanding, and GRUs with self-attention for contextual information.
• A manually annotated dataset of tweets was created with binary and ternary labels for depression detection.

Main Results:

• The SSCL framework achieved 97.4% accuracy on binary labeled data and 82.9% on ternary labeled data for depression detection.
• An F1-score of 94.4 was obtained on unseen random tweets.
• The framework demonstrated superior performance in cross-domain validation for sarcasm detection on news headlines.

Conclusions:

• The proposed SSCL framework effectively detects depression from textual data, outperforming existing methods.
• The model's ability to capture implicit context is a key strength.
• The framework shows promise for broader applications in mental health monitoring and analysis across different domains.