Deep Learning Techniques For Spamming And Cyberbullying Detection

Abstract

Social media allows people to exchange their views on different topics. However, some users post offensive comments on social media, which is known as cyberbullying, and some scam users by posting fake links in the comment section which adversely affects the user experience on those platforms. Thus regulation of contents in social media has become a growing need. The metrics for the regulations can be achieved by detecting these comments using deep learning techniques and machine learning algorithms Gaussian Naive Bayes, Logistic regression, Decision tree classifier, Adaboost classifier, Random forest classifier, MBERT and BERT, Gaussian Naive Bayes, Logistic Regression for cyberbullying and spamming respectively.This paper aims at detecting cyberbullying and spamming with the help of the techniques mentioned above for spamming and cyberbullying detection in English and Tanglish. Furthermore, it compares a number of supervised techniques, including standard and ensemble methods. Ensemble supervised methods have the capability to outperform conventional supervised methods, according to the evaluation of the results. The fine tuned classification BERT model (Bidirectional Encoder Representations from Transformers) and MBERT(Multilingual BERT) are based on Transformers, are used for cyberbullying detection and machine learning algorithms which include Gaussian Naive Bayes and Logistic regression have been used for spam detection. We do a comparative study between deep learning and machine learning algorithms. Compared to machine learning algorithms, deep learning algorithm BERT performed better at detecting spamming comments.