Hyper-Graph Attention Based Federated Learning Methods for Use in Mental Health Detection

Area of Science:

• Digital mental health
• Computational linguistics
• Machine learning in healthcare

Background:

• Internet-delivered psychological treatment (IDPT) requires robust AI models for accurate symptom detection.
• Deep neural networks (DNNs) need diverse data; models trained on limited datasets perform poorly in new settings.
• Improving data diversity and distinctiveness is crucial for reliable AI in mental healthcare.

Purpose of the Study:

• To develop a federated learning embedding model for mental health symptom detection using a novel structural hypergraph and emotional lexicon.
• To enhance word representation, enabling vocabulary diversification, grammatical analysis, and dynamic lexicon analysis.
• To create semantic word representations via an attention network model for clinical text analysis.

Main Methods:

• A structural hypergraph and emotional lexicon were proposed for word representation.
• An embedding model utilizing federated learning was developed for text-based symptom detection.
• Attention-based mechanisms and bidirectional LSTM architecture were employed and experimentally validated.

Main Results:

• The proposed strategy effectively addresses vocabulary diversification, grammatical representation, and lexicon analysis.
• Experimental results demonstrated the encoding of emotional words using the structural hypergraph.
• A 0.86 ROC was achieved using a bidirectional LSTM architecture with an attention mechanism.

Conclusions:

• The developed federated learning model with a structural hypergraph and emotional lexicon shows promise for accurate mental health symptom detection.
• This approach enhances the generalizability of AI models in IDPT by leveraging diverse and distinct data.
• The method offers a robust framework for semantic word representation and clinical text analysis in mental healthcare.