Leveraging Named Entity Recognition to Enhance Digital Mental Health Support for Domestic Violence Victims in the Post-Pandemic Era

Yuting Feng^*

• School of Materials Science and Engineering, Nanjing Institute of Technology, Nanjing, China

The growing demand for trauma-informed and scalable digital mental health services underscores the necessity of robust AI frameworks capable of interpreting complex psychological signals from irregular and multimodal behavioral data. This study introduces a novel architecture that integrates Named Entity Recognition (NER) with graph-based modeling to enhance mental state inference for survivors of domestic violence in post-pandemic contexts. Central to our approach is the Adaptive Contextual Psychological Graphs (ACPG), which constructs individualized psychological trajectories by encoding behavioral, contextual, and interpersonal dynamics. This is further supported by Recursive Psychological Inference Scheduling (RPIS), a meta-learned inference mechanism that dynamically adjusts prediction schedules based on uncertainty and structural alignment. Extensive experiments on four benchmark datasets—UCSD Pedestrian, MVTec AD, KDD Cup 1999, and NAB—demonstrate that our model consistently outperforms strong baselines, achieving up to 3–5% improvements in F1-score and AUC under low-supervision conditions. Ablation studies confirm the complementary roles of ACPG and RPIS in capturing fine-grained mental health indicators such as social withdrawal, emotional volatility, and dissociative behavior. The results highlight the feasibility of deploying AI-assisted, context-aware mental health monitoring systems, offering valuable tools for early intervention and risk assessment in public health and social support settings.