Natural Language Processing and Naïve Bayes Classifier Algorithm to Automate the Detection of Cyberbullying

Abstract

Abstract: The impact of social media on contemporary culture has been unprecedented, making it the most significant medium of our times. While it has had a positive effect on people's worldview, social media has also been linked to a rise in undesirable phenomena such as cyberbullying, cyberstalking, and cybercrime. Cyberbullying, in particular, can have a negative impact on individuals' mental health and has even been identified as the root cause of mental health issues in some cases. The proliferation of sexually explicit comments and the spread of rumors by multiple individuals are some of the negative influences that have been observed in the social media ecosystem. In recent years, academics have been increasingly concerned about the indicators of online harassment. Our goal is to develop a system that can detect instances of online abuse using Natural Language Processing (NLP) and Naïve Bayes, among other techniques. The cultural norms have shifted dramatically due to the rapid transmission of the COVID-19 virus, resulting in a rise in cyberbullying, especially among adolescents. The younger generation is more likely to engage in this practice, which has become more widespread with the stratospheric rise in popularity of various online engagement-promoting platforms. The COVID-19 pandemic has changed the way people interact online and has contributed to an increase in cyberbullying. As more people began working from home, bullying became a more significant concern. Our proposed system includes modules for data cleansing, text mining, word embedding, and regression analysis, among others. We utilize the Lemmatization technique for text mining, which enhances the model's precision. We also utilize the Vader emotion for feature extraction, which generates word vectors that are scattered numerical representations of word attributes. Additionally, Naive Bayes is used for data categorization to prevent overfitting in the proposed model. This would help in creating vectors that connect words with similar meanings