Multiple language Discovery Abuse of Cyberspace with Advanced Transformer Architecture

The number of reported instances of cyber abuse has also increased in tandem with the growth in the quantity of people that are regularly using the internet. Users' online freedom and privacy are jeopardized by such occurrences. Traditionally, the removal of such content from the internet has been accomplished through manual moderation and reporting systems. This strategy has, however, certain drawbacks, such as a reliance on people, longer delays, and less privacy for data. Traditional recurrent sequence models and supervised machine learning have been used in previous attempts to automate this procedure, but they have a tendency to perform badly on non-English material. A versatile solution that can handle multilingual text is urgently needed, given the increasing diversity of consumers using the internet. Moreover, translations frequently lose important context and emotion from content written in vernacular languages. In this work, we offer a generative deep learning based method for the detection of cyber abuse throughout Hindi, English with mixed codes, and Hindi text. The method uses bidirectional transformer-based BERT architecture.In the TRAC-1 standard aggressiveness identification task, the suggested architecture can get extremely positive outcomes on the task leaderboard for English and accomplish cutting-edge performance on the Hindi dataset with mixed codes. The outcomes show that the new strategy is superior to the current ones because they were obtained without the need of numerous models or ensemble-based methods. Effective models that utilize deep learning techniques to analyze multilingual text can process a wider variety of inputs, making them indispensable in combating these types of social ills.