Sexual Harassment Detection using Machine Learning and Deep Learning Techniques for Bangla Text

Abstract

Harassment is a kind of act that annoys or upsets someone. Harassment can be classified into different categories. Sexual harassment is one of them. Sexual harassment is a type of harassment that involves the use of implicit or explicit sexual overtones, including the inappropriate and unwelcome promises of rewards in exchange for sexual favors. At present time, the technology has become more advance and spread all over the place. That gave the toxic people a huge opportunity to spread toxicity in online platforms. Because of the increasing amount Bangla text in different social media platforms, we also need to filter such kinds of offensive Bangla texts. The objective of this research is to detect sexual harassment from Bangla text and classify them by using machine learning and deep learning algorithms as well as prevents them. In the experiment, we combined TF-IDF with different machine learning algorithms like Naive Bayes, Decision Tree, Random Forest, AdaBoost, SGD, Logistic Regression, KNN, SVM and got accuracy of 74.9%, 75.6%, 70.0%, 70.1%, 75.2%, 75.7%, 65.2%, 76.5% respectively. Deep learning algorithms like CNN, LSTM, hybrid CNN-LSTM were also used and achieved accuracy of 89% for all of them which is comparatively better than machine learning techniques.