MetaPromptMH

# MetaPromptMH: Context-Aware Mental Health Prediction via Adaptive Prompt Tuning **MetaPromptMH** is a context-aware and efficient prompt-based framework for mental health prediction on social media. It leverages meta-learning and soft prompt adaptation to improve generalization across diverse platforms and user communities, especially in low-resource and domain-shift settings. --- ## 🧠 Overview This repository provides the official implementation of the method described in: > **"Context-Aware Mental Health Prediction from Social Media via Meta-Learned Adaptive Prompt Tuning"** The proposed method combines: - **MetaPrompt**: A meta-learned prompt initialization strategy that enhances cross-domain generalization - **Context-Aware Prompt Fusion (CPF)**: A dynamic prompt selection module based on post-level and user-level signals - **Lightweight Adaptation**: Only prompt vectors are tuned, keeping the PLM backbone frozen --- ## 🌟 Key Features - 🚀 Efficient few-shot mental health detection with minimal task-specific tuning- 🌐 Robust generalization across platforms (Reddit, Twitter, etc.)- 📚 Compatible with BERT, RoBERTa, and other Transformer-based PLMs- ⚙️ Domain-aware and context-sensitive prompt fusion- 💡 Applicable to depression, anxiety, stress, and suicidal ideation detection tasks --- ## 📁 Project Structure ```bash.├── models/ # MetaPrompt and CPF implementation├── data/ # Preprocessed mental health datasets├── configs/ # YAML configuration files├── scripts/ # Training and evaluation scripts├── utils/ # Metrics, logging, and helpers├── checkpoints/ # Saved prompt encoders└── README.md