Racist and Abusive Memes Classification Using Deep Learning Techniques

Memes have grown significance as social media has grown in popularity. Memes are challenging to categorize using conventional techniques since they are typically constructed using a combination of images and text, and their content is frequently hilarious or sarcastic. We suggest a deep learning-based method for classifying memes into a variety of categories, such as sexism, politics, and criticism. This highlights the necessity for a system that can evaluate memes automatically before they raise controversy or spread humor. Before judging the text as racist and abusive or not, it will first extract the text from the supplied image. If the language is found to be unacceptable, the third phase will further categorize the information into three categories: mildly racist and abusive, very racist and abusive, and hateful racist and abusive. The five thousand memes that made up the dataset for this study were divided into four categories: hateful racist and abusive, highly racist and abusive, slightly racist and abusive, and not racist and abusive at all. In current history, there has been an upsurge in concern over the propagation of inappropriate memes on social media. The proposed project provides a deep learning-based approach to categorizing objectionable memes using CNNs, RoBERTa, and BERT. We enhanced pre-trained algorithms using a dataset of racist and abusive and non-racist and abusive memes. Utilizing various assessment metrics, like accuracy, precision, recall, and F1-score, we assessed each model's performance.