Hostile Social Media Harassment: A Machine Learning Framework for Filtering Anti-Female Jokes

Abstract

Drawing motivation from the non-existing computational criteria that blocks abusive contents targeted at female folks on social media from weird contents capable of hurting, offending, or intimidating the feelings; we present a machine learning framework for an intelligent filtering of information regarding anti-female jokes on social media discourse using multi-class classification. Our framework trains, validates and test neural network algorithm with data set sourced from chat rooms on social media handles with the goal of detecting and filtering expression(s) available in online jokes defined in terms of harsh, mild and neutral anti-female jokes. During implementation, the prototyped model recorded a performance accuracy of 92.9 %. This performance though high with the available limited data set is yet to attain the desired level of accuracy due to lack of consistent rules that distinguish various classes of anti-female jokes. In future, this model will be deployed to achieve unhindered discourse on social media by implementing an intelligent filtering system that could identify, classify and block unwanted contents.