ABSTRACT

Hate speech, which is a problem that affects everyone in the world, is taking on new dimensions and becoming more violent every day. The majority of people’s interest in social media has grown in recent years, particularly in the United States. Twitter placed 5th in social media usage figures in 2022, with an average of 340 million users globally, and human control of social media has become unfeasible as a result of this expansion. As a result, certain platforms leveraging deep learning approaches have been created for machine translation, word tagging, and language understanding. Different strategies are used to develop models that divide texts into categories in this way. The goal of this research is to create an effective a new hybrid prediction model that can recognize racist, xenophobic, and sexist comments published in English on Twitter, a popular social media platform, and provide efficient and accurate findings. 7.48 percent of the data were classified as racist, genderist, and xenophobic in the used dataset. A new hybrid LSTM Neural Network and Recurrent Neural Network based model was developed in this study and compared with the most popular supervised intelligent classification models such as Logistic Regression, Support Vector Machines, Naive Bayes, Random Forest, and K-Nearest Neighbors. The results of these several models were thoroughly examined, and the LSTM Neural Network model was found to have the best performance, with an accuracy rate of 95.20 percent, a recall value of 48.94 percent, a precision of 60.95 percent, and an F1 Score of 51.32 percent. The percentage of test data was then modified, and the comparison was made by attempting to get various findings. With a larger dataset, these deep learning models are believed to produce substantially better outcomes.

Keywords: - Artificial Intelligence, Deep Learning, Genderism, Racism, Xenophobia