Detection of homophobia and transphobia in YouTube comments

Abstract

AbstractUsers of online platforms have negative effects on their mental health as a direct result of the spread of abusive content across social media networks. Homophobia are terms that refer to the fear, hatred, discomfort, or suspicion of or toward those who identify as homosexual or bisexual. Transphobia is fear, hatred, discomfort toward those who are transgenders. Homophobia/transphobia speechs are a sort of offensive language that can be summed up as hate speech directed toward LGBTQ+ persons, and it has become an increasing concern in recent years. The homophobia and transphobia found online are a serious societal issue that can make online platforms toxic and unwelcoming to LGBTQ+ individuals and hinder the eradication of equality, diversity, and inclusion. We present a new dataset for online homophobia and transphobia detection that has been annotated by experts, which will enable homophobic and transphobic content to be automatically recognized. The dataset includes 15,141 annotated comments written in English, Tamil, and both Tamil and English. Additionally, we provide the outcomes of our benchmark system in a variety of machine learning models. For the purpose of developing benchmark systems, we conducted a number of experiments utilizing a variety of cutting-edge machine and deep learning models. Furthermore, we discuss our shared task conducted at LTEDI-ACL 2022 workshop to improve the research in homophobia and transphobia detection. It garnered 10 systems for the Tamil language, 13 systems for the English language, and 11 systems for the combination of Tamil and English languages. The best systems for Tamil, English, and Tamil–English each received an average macro F1 score of 0.570, 0.870, and 0.610, respectively.