Implementation of Machine Learning to Detect Hate Speech in Bangla Language

Abstract

Hate speech is a crime in all countries. Hate speech can be for women, religions, countries, cultures. The big problem for hate speech is that it entices the evil people. Moreover, it inspires them to spread hatred in the society. Bangla is one of the topmost spoken languages in the world. But hate speech detection in Bangla language is rare. Our purpose is to detect hate speech in Bangla language. To perform the task, we were in need of the Bangla datasets. But the Bangla dataset is not available. So, we have collected data from Facebook. Collecting data from the social site is very hectic. The data contain mixed languages, grammatical mistakes. So, we made a team to collect the data. Another team was to process the data. And finally, we labeled the data as hate speech or not. The team members had enough knowledge about hate speech. They were neutral towards the data. Our data contain hate speech against women, community, culture, ethnicity, race, sex, disability. Machine Learning approach is ideal for our work. We have used the SVM and Naïve Bayes algorithm for our work and got a maximum accuracy of 72%.