A Multimodal AI-Driven Framework for Post-COVID Digital Mental Health Solutions Targeting Domestic Violence Victims: Efficacy, Accessibility, and Ethical Considerations

Ming Zhang^*

• College of Geography and Resources, Sichuan Normal University, Chengdu, China

In the evolving field of digital mental health, there is a critical need for precision, interpretability, and ethical responsiveness, particularly when addressing the long-term psychological effects experienced by domestic violence survivors following large-scale public health crises. Traditional digital mental health interventions often rely on homogeneous sequence models and static symptom categorizations, lacking the nuance to handle episodic distress, trauma recurrence, or user-specific behavioral cues, and frequently fall short in ensuring transparency, adaptability, and ethical oversight. To address these limitations, we propose a multimodal AI-driven framework built around SympNet, a novel neural-symbolic architecture that models psychological state trajectories using heterogeneous behavioral data streams, including mobile interaction logs, physiological signals, and narrative disclosures. SympNet integrates temporal symbolic reasoning with attention-guided dynamic memory and latent manifold embedding to identify crisis escalation patterns while preserving semantic interpretability and respecting partial observability. We introduce Reflective Co-Adaptive Learning (ReCoL), an adaptive training strategy that combines expert feedback, user-specific cognitive rhythms, and symbolic safety boundaries to ensure context-sensitive updates and ethical stability. Our method is capable of real-time adaptation and zero-shot generalization, ensuring robustness even under data sparsity and emotional volatility. Empirical evaluations across longitudinal survivor datasets demonstrate superior forecasting of psychological deterioration, interpretable decision pathways, and increased clinical alignment compared to standard deep learning baselines. By embedding ethical awareness and clinical interpretability at the architectural and optimization levels, our framework lays a scalable, safe, and effective foundation for next-generation digital mental health solutions tailored to the complex needs of trauma survivors.