Classifying the Cyber Bullying Bengali Words From Social Media Comments Using Machine Learning Algortihms

Abstract

Social media, which provides numerous options for contact, but raises the hazard and frightening circumstances online, particularly for the young people worldwide. The key to effective mitigation is recognition system of potentially hazardous communications. Although much study has been done on English-language material, non-English material, particularly Bangla content, has traditionally been neglected. The use of subscriber data evaluation and machine learning techniques to address these difficulties in comments online or chats shows improved results. Many machines attempting to learn strategies are suggested in English language and literature. However, resolving the problem by using the approaches at hand is rather subject specific, therefore false identification might happen if the materials change from formal English to profanity or snark. Additionally, the political and social behavior of the research group and the linguistic variances between English and non-English material may affect performance. This study investigates how well-known machine learning techniques perform and how accurate they are when applied to Bangla text. Additionally, the influence of subscriber data, such as location, age, gender, the number of likes and comments, etc., is examined for bullying in Bangla. Support Vector Machine (SVM), which has an accuracy rate of 88.0% according to experimental data, is the most effective algorithm for detecting bullying in Bangla. However, when integrating subscriber data with user-posted data, KNN(3- Nearest) obtains the best accuracy with 77.8%, while in the same situation, Support Vector Machine (SVM) reaches 88% performance, which is quite near to the best one. Due to SVM's superior performance in both scenarios, it was decided to use SVM when applying the model to digital networking. As a consequence of the program, a java internet alternative has been created, and according to the reliability of the system, majority of the time the results are comparable to those of a distant analyst.