Mobile sensing and sentiment-based social media mining for surveillance anxiety in smart urban environments

The proliferation of smart city Surveillance Infrastructure (SI) has raised significant concerns about psychological impacts on urban populations, yet empirical measurement of Surveillance Anxiety (SA) in naturalistic outdoor environments remains limited. Existing approaches rely primarily on self-report surveys or single-modality behavioural analysis conducted under controlled conditions. This study presents the Multimodal Surveillance Anxiety Detection Framework (MSADF), integrating continuous mobile sensing with Social Media Sentiment Analysis (SMSA) for objective, population-scale SA detection in Smart Urban Environments (UE).

A longitudinal study was conducted across three urban districts of Coimbatore, Tamil Nadu, India (\(\:11^\circ\:^}^}\)N, \(\:76^\circ\:^}^}\)E) over eight months (March–October 2023), involving 325 enrolled participants — of whom 283 completed the full protocol — using custom Android applications for continuous behavioural monitoring (GPS trajectories, accelerometer data, ambient audio levels, and Bluetooth proximity detection), alongside 31,294 geo-located social media posts from Instagram and X (formerly Twitter). TCNs modeled behavioral sequences, and GCNs analyzed spatial distributions of sentiment across urban surveillance zones. A late-fusion model integrated behavioural and sentiment representations for final SA inference.

The MSADF achieved \(\:81.74\) accuracy and an F1-score of \( 81.82} \) (AUC-ROC \(\:=0.8967\)), surpassing the best single-modality method by \( 13.0} \)in relative terms. Spatial analysis revealed a significant positive correlation between SI density and anxiety levels (\(\:r=0.692\), \(\:p