Unmasking Hate: Deep Learning-Based Detection of Violent Incitement in Social Media

Abstract

The increasing usage of social media has amplified the spread of fake news as well as violent incitement content, causing critical concerns about societal harmony. Social media platforms are actively engaged in combating the dissemination of fake news around the clock; however, the overlooked content that incites violence, spreads stress, and fosters hatred has received insufficient attention. In connection, this study proposed machine learning techniques to detect and classify inciting violence in social media text. Leveraging the power of DistilBERT, a pre-trained transformer model, the research introduces a robust framework for identifying and classifying harmful content into various domains, including but not limited to political, religious, cultural, racist, and general categories. The dataset used underscores a versatile range of tweets, annotated for sentiment and contextual categories, allowing accurate examination of harmful communication patterns. The model’s result, illustrated through precision, recall, and F1 score metrics, outlines its productivity in taking convoluted, context-sensitive data. This innovative approach leads to keen insights into the dynamics of online incitement, providing a solution for content moderation and refining strategies. Aside from its technical results, this study also ensures the ethical concerns of filtering out media content and the societal requirement of automated systems to tackle digital hate expressions. These results allow policymakers, platform regulators, and researchers to refine content moderation practices and ensure safer online interactions. Future directions include the joining of multilingual datasets and filtering models for nuanced hate speech detection across vast cultural contexts. This research is a step towards leveraging artificial intelligence to mitigate the most pressing challenges in the digital age, incitement to violence, and instilling more inclusive online environments.