A Comparative Analysis of AI Models for Detecting Inappropriate Digital Media Dialogues

Billions of individuals around the world use social media networks every day. Social networks can be used for a lot of different things, from connecting with people in online communities to running a business or selling things online. But information on social networks can include violent and hateful speech that hurts others. Recent experiments have been undertaken to detect violent speech on social media in English and various other languages. Because hostile and abusive content on social media isn’t tied to a certain language, it can be dealt with utilizing AI methods, no matter how formal or casual the language is. The proposed approach focuses on brief English texts, primarily tweets disseminated on Twitter. The tweets in the dataset were sorted into two groups based on whether or not they included or spread hate speech. After that, LSTM, XGBoost, and BERT models were employed to do a binary classification. The BERT model was better than other models at predicting inappropriate tweets. It had the lowest false-positive error rate (2.6%), the highest F1 score (94%), and the highest accuracy (94.3%). The results from this study show a big improvement above other cutting-edge methods that have been published. This study’s implications and prospective research avenues are also examined.