HATE SPEECH DETECTION USING MACHINE LEARNING

Abstract

The evolvement of the Internet and social websites gave society multiple platforms where everyone has free rein to disseminate their ideas. However, this liberty is being abused by some to spread hate speech texts on these networking forums against particular individuals or circles of people due to their ethnicity, faith, gender, etc. In recent years, these hate speech texts could be perceived as the seeds for cyberbullying cases and major conflicts all across social media forums. Thus, to solve this concern of hate speech texts recent studies incorporated a majority of feature engineering approaches and machine learning models to automate this process of detecting hate speech in messages on various datasets. Hence, this paper aims to collate comparisons between numerous machine learning models and train a machine learning model using publicly available datasets giving the best accuracy and being the most capable model for detecting hate speech. This paper talks about the classification of audio data extracted from Videos that can be classified as either normal or offensive based on their spoken content. To create the video dataset, a web scraping program called Crawler is used to identify videos containing specific types of offensive content. After that audio is extracted from the video using a speech-to-text converter and converted into textual data. After that, we use this textual data and work with other machine learning algorithms.