An Artificial Intelligence Approach To Detect Hate Speech From Bangla Sentence

Abstract

The kind of speech that occurs on the internet is intended to target a person or group on the basis of their religion, ethnicity, gender, disability, or even the color of their skin. Facebook and YouTube are two of the most prominent social media platforms in Bangladesh, and both of them frequently feature talks of this kind. People who preach hatred sometimes do it in the comment sections of celebrities' websites and social media accounts. Because of the proliferation of hate speech in recent years in Bangladesh, there have also been incidences of religious violence and suicides in that country. The removal of negative content from social media platforms has made it necessary to filter out certain types of comments and viewpoints. Consequently, the identification of hate speech expressed in the Bangla language has been our primary objective. There were a few works that came before this one, but they were not satisfactory in any way. A dataset that has more than eight thousand comments gathered from various social media platforms is being used, making it one of the most extensive datasets ever utilized. We implemented the Support Vector Machine (SVM), Decision Tree, Random Forest, Logistic Regression, and K-Nearest Neighbor (KNN) algorithms in order to create a model that separates Bangla comments into normal speech and hate speech. The most reliable result in Bangla Language can be obtained by using our model, which is based on very specific calculations. Following an examination of the outcomes of each method, we selected the model that provided the highest degree of precision for the test data.