Recognition of Hate Speech using Advanced Learning Model-based Multi-Layered Approach (MLA)

Abstract

Hate speech becomes more complicated for the users of social media. Some users on online social networking sites (OSNS) create a lot of nonsense by uploading hate speech. OSNS applications developing many models to prevent this hate speech in terms of text and videos. However, these messages still need to be fixed for OSNS users. Sophisticated techniques must automatically identify and detect hate speech material to solve this problem. This paper proposes an advanced learning model-based Multi-Layered Approach (MLA) for hate speech recognition. The proposed model analyses textual data and finds hate speech patterns using multiple deep learning (DL) architectures. The algorithm can generalize well across settings and languages because it was trained on text datasets that include various hate speech types. The final step is an integrated model called Text Convolutional Neural Networks (TCNN), which combines hate text pattern detection with T-Convolutionals. Essential components of the model include the pre-trained model for DistilBERT, integrated pre-processing techniques like Text Cleaning, Lemmatization, and Stemming, and feature extraction techniques like GloVe and Bi-grams (2-grams) to capture contextual information and nuances within language. The model integrates continuous learning techniques to handle the dynamic nature of hate speech. It enables the model to update its comprehension of new language patterns and evolving forms of objectionable content. The evaluation of the proposed model involves benchmarking against existing hate speech detection methods, demonstrating superior precision, recall, and overall accuracy. Finally, the proposed MLA offers a practical and adaptable solution for recognizing hate speech, contributing to creating safer online environments.