Utilizing Topic Modelling To Identify Abusive Comments On YouTube

Online video platforms such as YouTube was once regarded as a haven for entertainment, educational, and promotional purposes. Now they have become a breeding ground for spreading toxic behavior, radicalizing content, and political propaganda. We have used YouTube data API v3 to scrape several data related to youtube videos like videos URL, description, view count, and comment information like commenters, comments, and replies. We investigated some hot topics which are prone to abusive comments. For example, bullies are highly existent in racial, teenage lifestyle and appearance, LGBTQ topics, or targeted mainly towards women and girls. We randomly selected a couple of videos from these YouTube channels and used them to research and analyze this project. We have used LDA (latent Dirichlet allocation) to identify dominant topics in a particular uploader’s comment section. We have also used the TextBlob library of python for determining the polarity and subjectivity of comments for each uploader.