Hate Tweet Extraction from Social Media Text Using Autoencoder Wrapped Multinomial Naive Bayes Classifier

Social media is one of the biggest source for gathering real-time data, and information spreading in day-to-day life. In the name of freedom of speech, many people are using derogatory words and offensive languages which create a lot of problem for the concern person, organization, society, and government. To reduce the dehumanizing activities arises due to these hate speeches, we need to filter them from the social media text. Here, we propose a system which analyzes the micro-blog text to classify them as hate or non-hate speech and filter out the hate speeches. Our presented prototype system also handles non-English text as the social media platforms are not confined to English only. Various classifiers along with several learning techniques are used to exhibit their effectiveness for an accomplished solution. An autoencoder wrapped Multinomial Naive Bayes (MNB) classifier is incorporated in the system for a significant gain in the performance. The simplicity in its implementation makes our system superior to other state-of-the-art systems available in the literature. Finally, the prototype system is evaluated on various publicly available hate-related datasets to showcase its effectiveness on this challenging task.