Detection of Offensive Language in Code-Mixed Text on Social Media

In today’s digitally interconnected landscape, people have started expressing their thoughts and views on social media platforms. Expressing such freedom of thought occurs not only in English but also in a combination of native and English languages. Such a combination is called a code-mixed language. Expression of thoughts has positive as well as negative impacts, such as a decrease in privacy, cyberbullying, isolation, polarization, comparison, envy, and so on, posing significant challenges for maintaining a positive online environment. One of the negative outcomes can be the use of language that is offensive or disrespectful. We developed a novel framework to indicate the proliferation of offensive language. For this framework, we introduce a code-mixed corpus (Tamil, English, and Tamil-English text) for the detection of the offensive text within the comments. This work is introduced to scrutinize the systemic relationships among comments, users, and their interactions, thus enabling precise detection of offensive language patterns. Unlike traditional models, this framework considers not only the content of individual comments but also the intricate interconnections within the comments. Addressing this issueis essential not just for the well-being of users but can also give a few other advantages on social media platforms. A pre-trained transformer-based model approach is proposed for the classification of offensive language in code-mixed text. This proposed method attained an F 1-score with an overall accuracy of $67.07 \%$.