SmartShield: A Hybrid Deep Learning-Based Chrome Extension for Real-Time Detection of Spam and Toxic Comments on YouTube

Abstract

The rapid proliferation of spam and toxic comments on video-sharing platforms like YouTube has become a significant threat to user safety and digital well-being. This paper proposes ”SmartShield,” a robust hybrid deep learning-based Chrome extension designed for real-time content moderation. The system integrates state-of-the-art architectures, including RoBERTa for deep contextual analysis and a hybrid CNN-LSTM network to capture sequential patterns and linguistic obfuscations. To ensure high reliability, an ensemble layer utilizing SVM and XGBoost is employed for final classification. Beyond mere detection, the framework incorporates Explainable AI (XAI) through SHAP and LIME, providing users with transparent insights into why specific comments were flagged. The extension operates seam- lessly at the browser level, utilizing a Flask-based backend and the YouTube Data API v3 for immediate intervention. Experimental evaluations demonstrate that this hybrid approach significantly outperforms standalone models, achieving superior accuracy and a reduced false-positive rate. This research provides a scalable solution to online harassment, empowering users with a proactive defense mechanism against evolving cyber threats The proposed hybrid model achieved an overall accuracy of 94.9%, with a precision of 95.2%, recall of 94.6%, and F1-score of 94.8%. Experimental evaluation demonstrates that the ensemble approach outperforms individual baseline models such as SVM, CNN-LSTM, and RoBERTa when used independently.