School of Engineering 300597 Master Project 1 Sydney City Session 2 2020 Edition: Sydney City Session 2 2020 Copyright ©2020 University Western Sydney trading as Western Sydney University ABN...

Contents
Abstract    2
List of Figures    3
List of Tables    4
Introduction    5
1.1    Background of the research    5
1.2    Research Rational    5
Research Aims and Objectives    6
1.3    Research Questions    6
1.4    Research Significance    6
Literature Review    7
2.1 Concept of Deep Learning    7
2.2 Concept of Prediction System in Deep Learning    7
2.5 Fake News    9
Methodology and Models    11
3.1 Fake news detection using Image Processing    11
3.2 Fake news detection based on linguistic features    16
3.3 Hate speech classification using convolutional networks    21
3.4 Hate Speech detection using Deep Learning    24
Conclusion    30
References    31
    
Abstract
The exponential growth of the web has helped empower the voices of individuals but abuse of speech has also contributed to a rise in the number of cyber-crimes and antisocial behaviors. Hate speech, cybe
ullying, fake news are some of the issues to take quite seriously as otherwise, it could put social networks at risk for their honesty. In daily media outlets such as social media feeds, news forums, and news organizations, the abundance of misleading information has made it difficult to recognize trustworthy news platforms and thus increase the need for analytical tools to provide insights into the reliability of online content. By spreading misinformation it plays an increasingly dominant position by influencing the perceptions or understanding of individuals so as to distort comprehension and choice. Digitization transforms human contact into an electronic medium that has many advantages but also offers an atmosphere for anti-social online behavior, such as abuse, insult, and other types of offensive text. The purpose of this paper is to review and highlight the work related to the topic of Deep learning for prediction systems such as Fake news, hate speech, and cybe
ullying. The work includes the literature review, methodologies adopted and the conclusions are drawn. The literature review presented in this research will give enough substance for the proper understanding of the keywords and its relevance. The well-defined research methodology will help in understanding the different algorithms and systems developed to tackle the above-mentioned problems. Towards the end, important conclusions are drawn out from these works and the scope of future work is also discussed. Keywords: Antisocial Online Behavior, Natural language processing, Text classification, Deep
Learning, Cybe
ullying, Attention mechanism, Hate Speech, Cybe
ullying, fake news detection, Pattern recognition.
List of Figures
    S. No.
    Description
    Page No.
    1.
    Suggested approaches to fake news detection
    
    2.
    Samples from the CASIA
    
    3.
    FAR/FRR results of ELA part of the system for various scale and quality parameters
    
    4.
    Hate-speech classifie
    
    5.
    LSTM topology for textual and acoustic model
    
    6.
    LSTM topology for multi-model model
    
    7.
    
    
List of Tables
    S. No.
    Description
    Page No.
    1.
    Sample legitimate and crowdsourced fake news in the Technology domain
    
    2.
    Sample legitimate and web fake news in the Cele
ity domain
    
    3.
    Classification results Fake News dataset collected via crowdsourcing
    
    4.
    Classification results for the Cele
ity news data set
    
    5.
    Agreement among two human annotators on the Fake News AMT and the Cele
ity datasets.
    
    6.
    Performance of two annotators (A1, A2) and the developed automatic system (Sys) on the fake news datasets
    
    7.
    Twitter hate-speech dataset statistics
    
    8.
    System performance (10-fold cross-validated)
    
    9.
    The number of collected data
    
    10.
    Experiment results using word embedding
    
    11.
    Experiment results using acoustic features
    
Introduction
1.1 Background of the research
Cybe
ullying, trolling, hate comments and fake news are some of the critical concerns of society. According to the digital world, more than 27% of the youth are affected and harassed because of the rise in cybe
ullying and telling. The use of the internet has created several damages to the minds of the child. More than 13% of the children have been found under clinical depression due to the ill effects of cybercrimes (Olweus & Limber, 2018). The Jump Tech Blog estimates for example that Facebook connections constitute 50 % of total falsified news website traffic and 20 % of the total flagship website traffic. Since the majority of U.S. adults –62%– gets news on social media (Jeffrey and Elisa, 2016), being able to identify fake content in online sources is a pressing need. Germany also decided that if it failed to delete illegal content on time, social media companies would face a fine of $60 million. Denmark and Canada have legislation that prohibits all speeches with abusive or offensive material directed at minorities and which could encourage violence and mental disorders. The Indian government has also called for the required action against hate speech, especially in posts that hurt religious sentiments and trigger social uproar, from prominent social media sites like Facebook and Twitter. The false information circulated on the internet is the result of various drastic changes undertaken by the youth. Therefore, the research study aims to provide specific technological solutions to
ing a decisive move against the ill effects of cybe
ullying.
1.2 Research Rational
The excessive increase of cybe
ullying, hate comments, trolls and fake news is the major research issue. The research addresses how the concepts of deep learning can help in mitigating the ill effects of the internet over society. It is an issue because of the excessive use of the internet has created a clinical depression among the youth of society. The misuse of the internet is affecting the lives of the people in many different aspects may it be good or bad. Every Individual is using social media platforms as they have become the central point to express yourself or to reveal someone’s opinions of thoughts. There are individuals who are misusing the technology to misguide and influence people, leading to riots or tensions between different religions, posing a threat to the integrity and unity, etc. The research sheds light on the necessity of deep learning to restrict the misuse of the internet. The study provides a new dimension towards discovering ways to limit the negative impacts of the internet.
Research Aims and Objectives
1. To understand the concepts of Deep learning
2. To understand the effectiveness of deep learning on restricting the issue of the internet.
3. To evaluate the various impacts that use of the internet has on the youth.
4. To understand how deep learning can be implemented in the provision of hate speech and trolling.
5. To evaluate how deep earning can
ing out a positive impact on the mindsets of the trollers and the misusers of the internet.
1.3 Research Questions
1. What are the concepts of deep learning?
2. How can in-depth learning help in restricting the misuse of the internet?
3. How can deep learning have a positive impact on the minds of the users of the internet?
1.4 Research Significance
The research is highly significant because it tries to find solutions to cybe
ullying, fake news, and other ill effects of the internet. The study is essential as it can prove to be a pathfinder towards the na
ow and conservative world of the internet. Through this research, the effectiveness of technological advancements would be understood. It shows how Deep Learning can be utilized to restrict the ill growth of misuse of the internet (Englander et al. 2018). The research is significant, as it would help in
inging a new dimension in the protection of legal privacy. The study would help in restricting the misuse of the internet.
Literature Review
2.1 Concept of Deep Learning
Deep learning is a
anch of artificial intelligence that is designed to imitate the human
ain and its working capabilities. Deep learning also termed as a deep neural network or deep neural learning is a type of automated learning in which computers get to know and understand the universe in terms of the hierarchy of concepts. Since the machine gathers knowledge, it is not required to officially define aspects of the human-machine operator. It is the methods, in which the processing of the data is possible to detect the objects, to recognize the speech, and in the translation of the languages. Deep learning makes it possible to learn representations for data with multiple levels of abstraction from the machine model consisting of multiple layers of computation. The state of the art in speech recognition, visual objects recognition, object detection, and many others such as drugs and genomics have been greatly enhanced with these methods. Deep learning exposes a complex structure in large data sets by using the backpropagation algorithm to show how a computer will change the internal parameters used to measure the display in each layer in the previous layer. Deep learning is created and processed in such a way that it can comprehensively analyst and take up the decision, just like the human can (Chang, 2018).
Applications including natural language processing, language recognition, computer vision, on-line suggestions, bioinformatics, and videogames are studied as a part of deep learning techniques used by industry professionals like profound feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling and realistic methodology. Deep learning has already proved useful across other areas, such as computer vision, audio and voice processing, natural language processing, robotics, bioinformatics and chemistry, video games, search engines. The networking system of deep learning is an integrated structure; each layer is comprehensive to be built on the previous layers; therefore, with added information regarding hosts, senders, and others.
2.2 Concept of Prediction System in Deep Learning
The prediction system is the term refe
ed to understand the processed algorithm. The algorithm is trained on the databases before acting as an output. The prediction system is well understood and the outcome of the forecast of the new data is similar to some extent. The algorithm of the prediction system is meant for the generation of the values for each recorded data. It helps model builders to get the most value and identifiable resources for data (Yan et al. 2018).
The prediction system is mostly utilized in deep learning when determining the next predicted action in any strategy. Prediction systems have become a crucial research field to promote reliability and productivity in the new manufacturing industry. Internet of Things and cyber manufacturing methods help capture vast amounts of sensor data, which then provides support and enables effective processing and analysis of data using deep learning. In the deep learning neural system, the prediction system helps in the allowances of the business to have the most accurate guesses for the outcomes that are related to the associated data. For example, Prediction systems have played a significant role in the planning, inventory management, and quality assurance of plant maintenance. It is the first step for companies to predict the tangible value of the business. The deep learning mechanisms help in the development of the specific huge cu
ency model of prediction to prevent losing the customers.
2.3 Hate Speech
Hate speech is the kind of statement that is directed to spread hatred among the people either for one particular person or for the entire community. The hate speech can be directed towards the person for the city that is based on the specific race, sex, orientation, and so on. The term hate speech is a debatable topic in the present day as it is often considered to have collided with freedom of speech. In many countries, the laws have identified hate speech to be anything that provokes violence in the country. That can be any kind of gesture, any type of writing format, or any type of conduct. The hate speech laws are divided into two forms: the first one intends to preserve order in public and the second one is to protect the dignity of humans. The internet has become the platform for spreading hate speech by extremist organizations. However, several rules are implemented to stop spreading such propagandas (Gamback & Sikdar, 2017).
2.4 Cyber Bullying
Cybe
ullying, which can also be termed as cyber harassment, is the kind of harassment that takes place using electronic media or the internet. It is also popularly termed as online bullying. In simpler words, cybe
ullying can be refe
ed to as the usage of technology to harass, torment, humiliate, threaten, or intimidate the target person. The expansion of the digital circle and digital technologies has mainly raised the platform for the bullies to be indulged in such activities (Lee & Shin, 2017). There are several platforms, from which cybe
ullying can be done; these are through social media, messaging apps, forums, email, and so on.
Any platform, in which the information can be shared, becomes the platform for cybe
ullying. Cybe
ullying can be of different types, such as doxing, harassment, cyberstalking, impersonation. When the information of the targeted individual is published online to defame and humiliating, it is termed as doxing. When the hurtful or threatening comment is sent to the victim, online or directly to them is termed as cyber harassing. To...