Sentiment Analysis of Digital Mental Health Narratives Among Domestic Violence Survivors Post-Pandemic

Heng Wei^*Yuze Zhang^*

• Shanxi University, Taiyuan, China

To tackle these issues, this research presents an all-encompassing computational system aimed at digital mental health narrative analysis for survivors of domestic abuse. The method integrates three key components: NeuroContextNet, a neural temporal-social encoder, PsyStrategy, a personalized strategy generator, and a formalized mental health modeling foundation grounded in cognitive-behavioral theory and affective computing. NeuroContextNet employs a dual-stream attention mechanism to capture both the temporal dynamics of mental states and the contextualsocial interactions that shape them. It processes multimodal behavioral data-including passive sensing, ecological momentary assessments, and digital text narratives-while leveraging graph-based embeddings to represent social influences. PsyStrategy builds upon these latent representations to generate adaptive micro-interventions using a psychologically-aligned policy engine. It utilizes inverse reinforcement learning, user receptivity modeling, and strategic diversity to produce personalized mental health strategies that are both effective and engaging. Moreover, the framework is designed with scalability and adaptability in mind, supporting dynamic integration with real-time data streams and personalized feedback loops. It not only accounts for the fluctuating emotional and psychological conditions of survivors but also aligns with clinical therapeutic goals. By embedding interpretability features and modular customization capabilities, the system provides actionable insights for mental health practitioners, while fostering long-term behavioral resilience in a digital therapeutic setting.