Enhancing Inclusive Online Conversations with Multimodal AI for Contextual Toxicity Analysis, Automated Rephrasing, and Bias Detection

Online communities and discussion platforms provide a suitable stage to share ideas, opinions and thoughts. But they have also created space to increase negativity by posting toxic and nuance contents and comments, leading to division, chaos and fights. These contents are sometimes targeted on a certain group of people based on gender, caste, religion and countries, paving ways to the disrupting of world peace. Some of these contents are targeted against an individual and are threatening and harsh, which can be coined as cyber bullying. The proposed system is aimed to provide a multi-featured solution that encompasses real time data moderation with the help of ensemble of embedder modules like GloVe, Emo2Vec and BERT to produce both semantic and contextual embeddings accompanied by BERT encoder and classifier, rephrasing module to provide inclusive paraphrases for the toxic inputs while preserving the information and a reinforcement module for the system to increase its adaptivity. This novel combination of modules overcomes the limitations of existing system by increasing the usability, robustness and real time processing.