Published Paper Details:

The extensive prevalence of social websites like YouTube has resulted in escalating cyberbullying and the ex- change of offensive information in user-added comments. Content moderation is an arduous and inefficient undertaking at scale, requiring automated platforms for monitoring. This paper illus- trates an AI system intended to filter and block cyberbullying online in real-time by monitoring comments on YouTube based on Natural Language Processing (NLP) and Deep Neural Networks (DNNs). The system makes use of the YouTube Data API to retrieve user comments in real time. Every comment is NLP-preprocessed with tokenization, stop-word filtering, and word embedding methods to prepare data for analysis. Objectionable content is identified through a combination of rule-based keyword matching using a pre-defined list of objectionable words and sentiment classification using a trained DNN model. The DNN model is trained to recognize the emotional tone and intent of user comments so that the system can mark not just overtly offensive language but also contextually offensive content. A warning system is implemented where a user is warned every time offensive content is detected. When the user gets three warnings, the system mimics an automatic blocking operation. Also, a web-based dashboard is created for administrators to track live comments, see flagged users, and monitor the number of warnings or blocks done. Experimental assessment demonstrates that the combined application of rule-based filtering and deep learning improves the accuracy and resilience of cyberbullying detection. The proposed framework can be extended to multiple social media sites and used for multilingual comment analysis. The system proposed is helpful in developing intelligent, scalable, and real-time moder- ation systems for safer online communication environments.