Research Journey of Hate Content Detection From Cyberspace

Introduction

In the 21st century, digital media provides scope for everyone to share information. People from all age groups are using the cyberspace every day. As per the survey from Statista (Clement, 2019), 100% of users of 18-29 age groups and more than 90% of the adult populations are using the internet in the United States in 2019. UK government presents national statistics (Prescott, 2019) where 99% of users of 16-44 age groups are using cyberspace. World Atlas survey shows (Dillinger, 2019) most internet users countries are China and India. Surveys indicate the effects of hate speech, cyberbullying, online extremism on social networking sites are very high. To protect the world from such negative situations, all tech giants are rigorously trying to improve automatic detection systems.

Hate speech is the action that abuse individual or groups based on various features like gender, colour, race, religion, nationality, and disability, etc. Every day numerous hate contents are posted by various users that are difficult to trace. Manual detection of hate content is laborious, time-consuming, and not scalable. Usually, manual approaches block websites or remove paragraphs that contain slur words. Social webs usually have a mechanism to report hate posts by user reviews, but automatic recognition of hate content is one of the priorities. During recent years, the importance of the automatic system of hate speech detection has grown. Various shortcomings still exist for automatic systems. Numerous posts circulating on the web are laborious for researchers to identify as a hate post.

Figure 1.

Hate texts from internet

The chapter significantly emphasizes automatic hate speech detection techniques and challenges. It also critically analyses the theoretical framework for hate speech as well as techniques to develop automatic hate speech detections systems. Different existing algorithms and feature representations performance in automatic systems are also part of this study. The majority of the study in hate-motivated speech classification relies on Natural Langue Processing (NLP) with various Machine Learning models. These have applied for binary classification and multi-class classification or both. Different supervised techniques (like SVM, Logistic regression, Decision Trees), unsupervised algorithms (k-means, bootstrapping) and deep learning methods (like CNN, LSTM, RNN) has employed to predict the hate posts. Evaluation measures section with various existing models helps to identify the progress of research.

The chapter aims to identify the most successful methods or techniques for hate speech detection to date. The chapter is structured with two fundamental parts related to the hate posts detection methods which help readers to gain knowledge for current Natural Language Processing Research. The theoretical framework and technical framework are the main part of this chapter that comes under the solution and recommendation section. The theoretical framework described with several related concepts and rules for hatred speech classification, whereas technical framework illustrated with various approaches implemented in the latest research. The next section will describe the research analysis of hate speech detection using the scientometric method. In the last section, the chapter ends with the future direction of hate content detection research.

Background

In today's world, cyberspace plays a very crucial role in our life. However, one negative side of the social web is cyberhate. Negative sentiments flowing in social media can potentially ruin social balance in society. Leading social webs describe hate speech as a violent speech that can be harmful to people. Online hate speech affects the youth of many countries, where they experienced severe depression. Automatic hate content detection research helps to balance social harmony. The automatic system is highly essential to protect cyberspace from hate speech effects. To shield society from hate speech, many countries are considered this as an illegal address.