Identification of Hate Speech and Offensive Content using BI-GRU-LSTM-CNN Model

Abstract

The propagation of hateful speech on social media has increased in past few years, creating an urgent need for strong counter-measures. Governments, corporations, and scholars have all made considerable investments in these measurements. Hate speech on social media platforms can lead to cyber-conflict that can impact social life at the individual and national levels. It can make people feel isolated, anxious and fearful. It can also lead to hate crimes. However, social media platforms are not able to monitor all content posted by users. This is why there is a need for automated identification of hate speech. The English text is notorious for its difficulty, complexity and lack of resources. When examining each class individually, it should be noticed that a many hateful tweets have been misclassified. As a result, it is advised to further examine the forecasts and mistakes to obtain additional understanding on the misclassification. To automatically detect hate speech in social media data, we propose a NLP model that blends convolutional and recurrent layers. Using the proposed model, we were able to identify occurrences of hate on the test dataset. According to our research, doing so could considerably raise test scores. Proposed model uses a deep learning technique based on the Bi-GRU-LSTM-CNN classifier with an accuracy of 77.16%.