Promoting Positive Discourse: Advancing AI-Powered Content Moderation with Explainability and User Rephrasing

Nothing is good or bad only thinking makes it so”- is a well-known phrase we might have heard while growing up. Social media and discussion platforms provide unsupervised and an unbound stage to share and discuss ideas, opinions and thoughts. Apart from these they knowingly or unknowingly pave way to a space that generates and harbors toxicity. The contents of these are sometimes targeted on a certain group of people based on gender, caste, religion and countries not keeping in mind the civic sense. This paper introduces an innovative AI-powered system for content moderation, designed to combat online toxicity. The system utilizes a rule-table-based filter for efficient initial screening, blocking content with clear violations. For nuanced toxicity detection, we employ an ensemble of RoBERTa and BiLSTM models. Promoting collaboration, we integrate GPT-3.5 to generate rephrasing suggestions, empow- ering users to modify their language constructively. The system incorporates explainable AI to clarify model decisions, enhancing transparency and understanding. Finally, a user feedback loop ensures the system’s ongoing evolution, maintaining its relevance to community standards.