© 2020 JETIR February 2020, Volume 7, Issue 2 www.jetir.org (ISSN-2349-5162) JETIR2002083 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 558 Detecting Hateful Content on Social Media Aaniya Gouse 1*, Afaq Alam Khan 1 1 Department of IT, Central University of Kashmir. India. Abstract With the growth of hateful content all over the web, detecting hatred has gained utter importance. To combat the nuisance self-regulatory methods are found to be in place but mostly these fail to serve the purpose. In this paper we have addressed the issue by training a supervised classifier that is trained based on semantic features. Just as semantic features work well with other experiments regarding sentiment analysis, this work is outperforms the state-of-the-art methods. Observing the performance of the classifier, we incorporate additional features in order that the performance of the classifier is maximized. Keywords: Social media, Natural Language Processing, Hate Speech, Semantic Features, Supervised Learning. Introduction The term hate speech does not have a clearly defined boundary. It stands practically undefined and its loosely defined boundaries are often crept over by “free speech”. For instance, calling a person a name could cleanly be categorized as being “free speech” but at the same time it could be full of hatred innately. Various aspects of hate are cyber bullying [Hosseinmardi et al (2015)], abuse [Nobata etal (2018)], flaming [Nitin et al (2012)], toxicity [Jigsaw (2018)] etc. Time and again every social media user has faced hatred online in forms varying from threat to abuse and so forth. Hateful language has an immense contribution to users abstaining from using social media altogether. Many platforms implement self-regulatory methods to keep a check on hateful content being propagated on the social media which include a user purposely reporting a particular profile as being offensive or violating certain guidelines. These methods being completely dependent over users’ discretion and their own definitions of hate are under qualified to be banked upon. Therefore, as communication online grows a need for an automated hate detector is ever increasing. Prospectively, our hate speech classifier shall prevent all hate crime while still arising. The goal of this research is to combat genocide, suicide, cyber bullying, trolling, terrorist propaganda etc. Our challenge is to detect hate out of the ulterior faces put on by it in the form of sarcasm, offense or misspellings. The state-of-the-art methods mainly employ statistical features but these do not lead to high accuracy and are error-prone as well. As semantic features in other classification tasks are observed to perform better we decided to make use of the same, making ours a one-of-a-kind hate speech classifier. In addition to semantic features, we shall incorporate lexical, morphological and contextual in order that the classifier turns out to be more intelligent. Contribution of Author The research is about designing a multi-class general-purpose classifier that is capable of detecting hateful content on social media. Specifically, our contribution can be stated as follows:  An efficient method for pre-processing of text which includes correcting misspellings and eradicating slangs.  A classifier that is capable of classifying hatred out of a given corpus.  Linguistic analysis of text to reveal syntactic and semantic details of language.  Evaluation of classifier on datasets of varying sizes and types in order to observe variations in performances.