Global Sentiment Analysis of Video Surveillance Technology Using BERT

Abstract

The significant imbalance between population growth and the availability of video surveillance systems in several major cities across different countries has raised concerns regarding the effectiveness of public space monitoring. As urban areas become increasingly dense and complex, CCTV is expected to function not only as a security support tool but also as part of a broader strategy for crime prevention, public safety, and urban management. However, the implementation of video surveillance also generates diverse public opinions, especially on social media platforms such as Twitter (X), where users actively express their views, support, concerns, and criticism. This study aims to identify public sentiment trends toward video surveillance and evaluate the performance of the BERT (Bidirectional Encoder Representations from Transformers) model in classifying these sentiments. The research uses social media data collected from various countries and applies BERT as a contextual natural language processing model. The findings show that most users expressed positive sentiments toward video surveillance, indicating that CCTV is generally perceived as beneficial for improving security and monitoring public spaces. In terms of model performance, BERT achieved its highest accuracy of 0.85 during the third trial at epoch 15. However, a slight decrease in accuracy occurred at epoch 20, indicating the possibility of overfitting when the model was trained for too long. These findings suggest that BERT is effective in capturing public opinion contextually and can be used as a valuable analytical tool to support evidence-based decision-making related to surveillance technology implementation in urban environments.