DEVELOPING A BROWSER EXTENSION FOR HARMFUL CONTENT DETECTION,

Abstract

Nowadays, the freedom of speech in cyberspace allows people to expresstheiropinions freely and easily. However, it also brings with it the risks of toxicandhatespeech, which can hurt people and affect their lives and cyberspace life ingeneral.Therefore, a tool to automatically detect toxic speech is very crucial toensureahealthy online environment. For that reason, this study presents the design and implementation of abrowserextension capable of identifying harmful content directly within web pages. Thesystem integrates client-side text extraction, preprocessing, and anoptimizedmachine-learning inference pipeline to achieve real-time detection without significantperformance degradation. A comparative study of Vietnamese language modelsisconducted to determine the most effective approach for browser-side deployment.Evaluation is performed on ViHSD dataset, measuring precision, recall, andresourceusage in Google Chrome browser.