Bangla Hate Speech Detection by Machine Learning

Abstract

The type of speech that takes place online intends to attack an individual or a group based on religion, ethnicity, gender, disability and even based on the color of their skin. Some popular social media in Bangladesh notably Facebook and Youtube filled with these types of speeches. The comment section of celebrities can be a perfect example of people spreading hatred. In recent times we can also see religious clashes and cases of suicide in Bangladesh because of the spread of hate speech. Filtering these types of comments and opinions has become a need to make social media free of negativity. So detecting hate speech in the Bangla language has been our primary goal. There had been a few previous works, but they were not up to the mark. A significantly large dataset is being used consisting of more than eight thousand comments collected from different social media platforms. We Introduced a model that classifies Bangla comments into normal speech and hates speech by implementing Support Vector Machine (SVM), Decision Tree, Random Forest, Logistic Regression, and K-Nearest Neighbor(KNN) algorithms. Our model with the help of specific calculation provides the most dependable result in Bangla Language. After analysis results of all algorithm, we choose the best model that is produced best accuracy for test data.