Please Be Nice: A Deep Learning Based Approach to Content Moderation of Internet Memes

Abstract

The internet has grown throughout time from a means for information distribution to a multipurpose tool that allows people to communicate, meet, and interact with one another. In socializing online, sending images with cultural and relatable context, otherwise known as memes, has become a common practice. Memes are sent everywhere, from forums and posts to even private messages. They are famous for converging thoughts, current events, or symbols creatively and colloquially. It combines text with a visual graphic and conveys many emotions, including hatred, which can harm other netizens. As a tool for communication, it can be used to convey humor or even as a tool for cyberbullying through hateful speech. Since memes are sent in image form, filtering out hateful speech in the form of memes can be difficult. To address this rising issue, researchers explore the use of deep learning approaches to detect hate speech on multimodal memes as a systematic approach for eliminating hate speech. The model performed well in this investigation, with an overall accuracy of 62.5%. When identifying pictures, it also has a high proportion of precision (71.37%). This brings the suggested model’s ability to detect hate speech in memes to a close.