Artificial Intelligence-Based Textual Cyberbullying Detection for Twitter Data Analysis in Cloud-Based Internet of Things


The instances of Internet harassment have become higher in the cyber world and can occur anytime during conversations in which people see, participate, or share content. Torture or provocation can be perceived as “sad behavior” with an intention of harming others. The occurrence of cybercrime in cyberspace includes erroneous messages, images, sounds, and videos that undermine or annoy [1]. Digital security is primarily aimed at young people and adolescents as it is the most dynamic form of meeting, using innovation, for various purposes such as participation, socialization, and so on. Cyberbullying involves sending malicious messages, defamatory comments, or images through unofficial communications, mail, and so on. The goals behind these are basically the parts of the web: tireless, the ability to watch and copy, just like the detectable crowd. Since the electronic exclusion significantly affects the society [2], it is highly important to consider this research field as the need of the hour [3]. Although this is a variant of harassment that occurs in offline world, the way it is done online has its own characteristics. For example, after the presentation of a demonstration on Internet, it could remain open on the Internet forever and not just breaking the normal barriers required for harassment [4]. Three strategies, such as the keyword system, mining estimation, and Internet life test are used in the methodology. These methods are combined as tests to deal with cyberbullying [5].

The online competition is divided into an initial section in which a short electronic avoidance database and all its basic procedures are provided following which a description of machine learning (ML) techniques is given. ML methods are used to identify the business on Internet. ML is characterized as the ability of a computer on decision-making with the help of accessible information and meetings [1]. The first step in understanding and preventing electronic blocking is its identification, and here we are likely to present potentially unsafe e-mails. These messages create particular difficulties for normal Natural Language Processing (NLP) as these messages are strange and full of spelling mistakes and outlines [6]. Percussion recognition strategies can be divided into two categories: one by slogan and another by Artificial Intelligence (AI). The easiest way is password technique that uses slogans to search for a sensitive substance in content. Although the thinking is basic, this technique is able to achieve high precision score using both search terms and web search [7]. The AI method is getting increasingly entangled. The three main sections of AI are representation, evaluation, and learning and these three comparative rules are designed for e-commerce recognition techniques. In the recognition of content based on impact circuit, the fundamental and foremost progress is the numerical representation of learning for immediate messages [8]. Indeed, the representation of content learning has largely focused on content extraction, data recovery, and NLP. Cyberbullying and cybercrime are interesting online topics to explore in detail [9].

Cyberbullying types

FIGURE 10.1 Cyberbullying types.

A huge taskforce focuses on identification and non-observation techniques to locate cyberbullying with the help of remarkable data from sexually explicit data, UI, and etymological and non-verbal references to realistic properties [10]. The discovery of digital death is considered as critical thinking and another calculation is suggested that aims at reducing the time needed to run an electronic promotion notice, thus limiting the amount of estimates required to select options. The planned position in cyberspace requires computational methodologies that can benefit from different properties of both vocal and non-verbal types [11]. In addition, several mechanical learning models are considered, including direct models, tree-based models, and deep learning models with numerous labels printed so as to make them an ideal model. Taking the collected tweets into account, the nature of planned discovery in digital madness is appreciated while the best model identified can reach an accuracy of over 90% with expected content expectations [12].

The levels of harassment have increased in India in manifolds. A total of 79% of Indians are aware and concerned about cyberbullying compared to 54% across the globe. A total of 53% Indians underwent cyberbullying, compared to normal 37% worldwide [13]. Likewise, half of the Indian population has involved in Internet harassment and only 24% of the total population has been confronted. In particular, the word Net database is used to identify semantically related words and to assess the proximity to selected cyberbullying terms. Again, a regulation-based methodology is proposed in the current study to identify the real cases of cyberbullying [14].

This chapter is organized as follows. Section 10.1 states the introduction phase in cyberbullying. Section 10.2 discusses the related works in cyberbullying. Then, Section 10.3 describes the proposed methodology in detail. Section 10.4 reports the experiments along with their results. Section 10.5 discusses the reported results, concludes, and summarizes this research chapter.

< Prev   CONTENTS   Source   Next >