Automatic Detection of Online Hate Speech Against Women Using Voting Classifier

Abstract

AbstractFreedom of expression found on social media has various pros and cons. Gender-Based Violence (GBV) is also a major issue in social media. As a part of GBV, hate speech against women is on the rise on all social media. There are some lapses available in the stand-alone classifiers in detecting such speech, and the performance of ensemble classifiers is much better. Also, many research works have focused on common hate speech datasets. Hate speech against women has been used in very few research activities. But such hate speech is very dangerous. As a result, this research employs, Decision Tree (DT), Logistic Regression (LR), Random Forest (RF), and Long Short-Term Memory (LSTM) to compute metrics and performances and then use those algorithms to create a voting classifier to develop a more accurate model for detecting hate speech against at women. Two phases were used in this study. RF, LR, DT, and LSTM were used as foundation stand-alone classifiers in the first phase of the ensemble procedure. In Phase Two, the weights of the second-level classifier were estimated using first-level classifiers. Hate speech against women was detected using an open-source #MeToo dataset that was utilized for training and testing by the researchers. The dataset is publicly available on GitHub which was uploaded by Nazmus Sakib. This dataset consists of 278,765 #MeToo movement posts on social media. It clearly shows that the proposed voting classifier model has the highest values in all metrics including accuracy (89%). When we check the strongly positive classification, the proposed model has performed well in precision (0.90), recall (0.91), and F-measures (0.90) and it can calculate strong positive hate speech more efficiently than other stand-alone classifiers. This voting model takes more time to train since it has multiple models inside. By training it for more epochs, we can further increase accuracy.KeywordsGender-based violenceHate speechWomenMachine learningDeep learningAnd voting classifier