Authors: Dr. Mainka Saharan, Mainka Saharan, Prince Kumar, Anuj Sharma, Sonu Kashyap, Yash Saxsenad
Abstract: Hate speech on social media has become a critical issue, posing a threat to societal harmony and individual well-being. As online platforms have become integral to communication, the dissemination of hateful and offensive language is increasingly unchecked, necessitating automated systems to detect and mitigate its impact [1][3]. This project aims to develop an automated hate speech detection system using advanced deep learning techniques, specifically the DistilBERT model, a lightweight transformer architecture known for its efficiency and accuracy [2][9]. The system categorizes textual content into three distinct classes: hate speech, offensive language, and neutral speech [1][4]. By employing comprehensive preprocessing methods to clean the text and leveraging tokenization to capture semantic meaning [1][6], the model is fine-tuned on a labeled dataset and achieves a test accuracy of 90.5%. The proposed system is designed for scalability and real-time deployment, addressing the challenge of moderating the vast amount of user-generated content on social media [5]. This study highlights the importance of using robust transformer models to analyze linguistic nuances, ensuring accurate classification even in complex and implicit cases of hate speech [9][2]. The project’s contributions include the development of a deployable application, introduction of data balancing techniques, and an evaluation of various preprocessing and modeling approaches [1][4].
DOI: http://doi.org/
