A Bidirectional Long Short-Term Memory Network and Bidirectional Encoder Representations from Transformers for Cyber-Bullying Detection

Social media has brought people from all around the world to one place to share their feelings, thoughts, and emotions with their close ones. However, using social media Cyberbullying involves harming social masses emotionally and sharing threatening content. Further, social media has become the spotlight of cyberbullying to down someone emotionally or physically with the use of aggressive words, abusive language, or even by the user of power. Machine and Deep learning models outperform to detect of cyberbullying, We Utilized Multiple Algorithms such as Naïve Bayes, Bi-directional long short-term memory (Bi-LSTM), and BERT (Bi-Directional Encoder Representation from Transformers) models, to detect bullying tweets using the pre-trained model of word2vec embeddings. This work efficiently reveals the techniques for the detection of cyberbullying over social platforms by classifying tweets for cyberbullying into five different classes: 1-Racism, 2-Age, 3-Gender, 3-Ehtnicity, 4-Not Bullying, 5-Other Bullying. The work in this paper will identify and provide cyberbullying in text, videos, pictures, and audio as an automatic tool to detect bullying and prevent the public from bullying using social media. We evaluate the model through a baseline model using the naive Bayes on the Twitter dataset. The baseline model compared with the proposed Bi-Directional LSTM using the same dataset which surpasses the baseline and achieves higher accuracy