Cyberbullying Detection using Machine Learning from Social Media comments in Bangla Language

Abstract

Cyberbullying which is defined as bullying perpetrated through the use of informa tion and communication technology is a serious problem nowadays. As a result of the invention of social networks friendships through different social media, relation ships, and social communications have all gone to a new level with new definitions. In fact, people become friends with someone whom he/she cannot even know face to face. With such a huge amount of users on the internet, cyberbullying has become a widespread global phenomenon. It not only makes a person mentally low but also has become one of the most important reasons for committing suicide. Being the seventh most speaking language in the world and increasing usage of the online platform, Bangla speaking people badly need an effective cyberbullying detection to handle this issue. In this thesis paper, we explore the spread of cyberbullying in fluence through the pairwise interactions between users. For cyberbullying through language, we will collect users’ unique comments from social media and check them with the help of psychological references. After that, those comments will be cat egorized using Word embedding, an evaluation tool to categorize text, so that the dataset will be shortened and ready for classification. Lastly, the dataset will be to a machine learning classifier named Random Forest in detecting the cyberbullying comments. The performance and accuracy of numerous frequently used machine learning approaches on Bangla text are investigated in this study. In addition, the influence of user-specific information, such as location, age, gender, number of likes, number of comments, and so on, is examined for the identification of Bangla cy berbullying. Random Forest is the top effective algorithm for Bangla cyberbullying identification when just posts or comments are used to identify, according to exper imental data, with 95.78% accuracy. Therefore, Random Forest is used for applying the approach on social media since it works better.