MindScope-AI

# 🌿 MindScope: AI-Driven Digital Mindfulness for Chronic Care **MindScope** is a video-based digital mindfulness framework designed to enhance engagement and attention in chronic disease management. It combines cognitive modeling, machine learning, and real-time nudging to deliver personalized, context-aware mindfulness support in primary care settings. ## 🧠 Motivation Managing chronic conditions such as diabetes and hypertension often requires sustained behavioral change. However, traditional mindfulness interventions face challenges like poor engagement, limited personalization, and high dropout rates. **MindScope** addresses these gaps by: - Detecting user attention in real-time from digital signals- Delivering tailored reflective prompts using counterfactual reasoning- Embedding mindfulness seamlessly into digital health workflows ## 💡 Core Components ### 1. Attentive Interaction State Encoder (AISE) A deep learning encoder that models user attentional states using multimodal digital interaction signals:- Action, context, and physiological input- Temporal embedding + attention-weighted modeling- Robust latent state stabilization with volatility-aware noise injection 📍 *Diagram and architecture explained on page 6, Figure 1 of the paper* ### 2. Reflective Nudging via Temporal Counterfactuals (RNTC) A reinforcement-learning-inspired mechanism to:- Predict when users are cognitively drifting- Simulate alternative intervention outcomes (counterfactuals)- Generate minimal but effective nudges using utility-based policies 📍 *Illustrated in Figure 3 on page 8* ## 📊 Experimental Results MindScope outperforms state-of-the-art video understanding models (e.g., CLIP, TimeSformer, Uniformer) across four NER datasets in terms of: | Dataset | F1 Score | AUC ||----------------|----------|---------|| CoNLL-2003 | 90.15% | 93.06% || OntoNotes 5.0 | 89.84% | 93.28% || WNUT-17 | 87.15% | 90.28% || GMB | 88.12% | 91.03% | 📍 *See Tables 1–2 on pages 11–12* ## 🔍 Ablation Study To validate component contributions:- Removing **AISE multimodal representation** causes a ~2.5% F1 drop- Excluding **contextual attention** reduces recall- Without **latent state stabilization**, robustness decreases in noisy settings 📍 *See Tables 3–4 on page 12* ## 🛠️ Installation ```bashgit clone https://github.com/yourname/mindscope-mindfulnesscd mindscope-mindfulnesspip install -r requirements.txt