Comparative Analysis and Assesment on Different Hate Speech Detection Learning Techniques

Abstract

Increased social networking services have modified the way and scale of cyberspace communication. Even so, owing to the anonymity and mobility of these services, the no. of online hate speech is growing. In recent years, several groups, scholars, and social media networks have worked to mitigate the detrimental effect of hate speech on social media. Despite these attempts, social media users continue to be subjected to hate speech. The issue is significantly more visible in social organizations that foster public discourse. Because it is costly and time-consuming to automatically identify hate speech by human annotators, an algorithm is needed for automatic recognition. It has proved effective to transfer knowledge with the finished of a pretrained language model for various downstream tasks in the area of natural language processing (NLP). This paper presents a deep analysis of how hate speech is detected by state-of-art speech detection methods. There are various methods to perform multitask learning in deep learning.The deep multi-task learning system to leverage valuable knowledge from these multiple tasks in numerous benchmark data sets showthe effectiveness of the method which is convincing about the most advanced models. Such DL techniques are used to detect hate speeches from the benchmark datasets. Also, a comparative evaluation has also done using these different types of machine learning techniques that also included deep learning with word embedding methods for hate speech detection. From this comparative analysis and discussion found that machine learning techniques achieved good outcomes but Deep learning techniques beat them. However, deep learning with word embedding techniques has achieved more than 92% performance using Bi-GRU-GloVebut deep learning without word embedding methods outperforms and achieved more than 95% performance efficiency results using LSTM over state-of-art techniques.