Blockchain Leveraged Cyberbullying Preventing framework

5 The popularity of social media has exploded worldwide over the last few decades and becomes the most preferred mode of 6 social interaction. The internet also provides a new platform through which adolescents are being bullied. Appropriate means of 7 cyberbullying detection is still partial and in some cases very limited. Moreover, research on cyberbullying detection extensively 8 focuses on surveys and its psychological impacts on victims. However, prevention has not been widely addressed. To bridge 9 the gap, this paper aims to detect cyberbullying efﬁciently. This paper employs a standard machine learning method and natural 10 language processing technique as a part of the detection process in decentralized Blockchain leveraged architecture. We provide a 11 fog based architecture for cyberbullying detection, aiming at relieving the server’s load by placing the detection and the prevention 12 of cyberbullying processes at the fog layer. The proposal might offer a probable solution to save users, particularly adolescents 13 from severe consequences of cyberbullying.

Section 2 reviews related work. We describe our architecture and method in Section 3. Finally, we present high-level 54 performance analysis, implementation of the model on a private Blockchain and accuracy of the detection method in Section 55 4 following with a conclusion. 56 II. RELATED WORK 57 An online social network allows a user to stay connected with friends and relatives more easily. People enrich their 58 relationships by sharing daily events and important moments. The way of sharing information is more open and details in 59 adolescents. Some approaches have been proposed to tackle online bullying. The paper [17] shows that the aggressive behavior 60 of anonymous users is more potential to lead cyberbullying. A different approach was suggested in 2012 by Chen et al. [4]. 61 Specifically, to predict the probability of to post offensive content, they included user's writing style, structure and specific 62 bully contents as a feature and introduced Lexical Syntactic Feature-based (LSF) model in the detection of online cyberbullying 63 from person's post pattern. The paper [19] proposed a data mining methodologies are applied to detect cyberbullying time 64 series modeling that identifies predator tactic in bullying which is determined by previous the state. This method also considers 65 bag-of-words, emotions, slang and abbreviations, a dataset of real-world conversation, and a numeric label is used to indicate 66 the severity of the predator's comments. Zhao and Mao [32] proposed a semantic-enhanced marginalized auto-encoder method 67 to address hidden cyberbullying feature in Twitter and MySpace data. Many researchers proposed a machine learning approach 68 to identify negative words and to learn language patterns between predators and victims with the ultimate goal of detecting 69 the possible occurrence of cyberbullying [28]. Xu et al. [29] proposed NLP methods to investigate questions (which text   Although solutions to prevent bullying include parents' effort, educators and law enforcement, but a technological advance-98 ment in filtering is not incorporated with current research. Nowadays, the attempt to stop bullying is now an emerging issue 99 to save adolescent from the severe consequences of cyberbullying. The existing research has not focused on the development 100 of a decentralized secure architecture for detecting and preventing cyberbullying. Blockchain technology has been successfully 101 introduced in cryptocurrency to ensure high security and privacy. Further, Blockchain has been explored to use in healthcare, Most of the conventional cyberbullying architecture comprises server and client. In General, bullying prevention or detection method is installed on a server or a client. The disadvantages of such a paradigm are that users might experience poor 109 QoS(Quality of Service). The server or client requires to keep the bullying prevention module active all the time which 110 demands higher power consumption in the server or client. Further, the bullying prevention method installed on the server or 111 client is often vulnerable to cyber-attacks including Ransomware, Denial of Service(DoS), and malware. To overcome these 112 shortcomings, we proposed a tier-based secure cyberbullying preventing architecture. The architecture has three tiers where 113 the bottom tier includes user's devices including smartphone, laptop, and computer, the middle tier consists of Edge or Fog 114 devices and the upper-tier includes traditional servers or Cloud servers. Fog [14] has recently brought computing resources 115 closest to user's devices. In Fog technology, computing capacities such as storage, processing are provided to a conventional 116 router, switch, and Gateway that is deployed close to IoT(Internet of Things) devices. User's devices send data to Fog node 117 which performs pre-processing before transferring the data or results to the Cloud server for further processing. Cloud services 118 are often migrated to Fog devices to provide users with QoS such as low latency. In our proposal, the bullying prevention 119 applications reside on the middle tier's devices to have faster processing of the Edge or Fog computing. 120 But, the bullying detection and prevention applications on a particular Fog device can be attacked and compromised by 121 malware or other cyber rogues or malicious users. Ensuring appropriate security and privacy at Fog devices is challenging 122 because diverse stakeholders belong to Fog devices and often remain unattended. Besides, heterogeneous Fog devices are 123 attributed to diverse security protocols and standards. Therefore, there needs a common mechanism to prevent malicious 124 attackers from modifying or stopping bullying prevention modules on the Edge network. Blockchain [24] has recently emerged 125 as a promising technology to withstand major security attacks. The Blockchain runs on a peer to peer network and maintains 126 a tamper-proof ledger replicated among each node that participates in mining. A chunk of data also called Block is added to 127 the Blockchain after a majority of Blockchain's nodes reach an agreement that the new Block is formatted following a suite of 128 Blockchain's rules. This process is called a consensus protocol. Once a Block is validated and then included in the Blockchain 129 network, the Block cannot be modified or changed further. A smart contract that represents a set of rules encoded by any 130 programming language is stored in every Blockchain node. The smart contract is triggered when a particular event related to that 131 contract occurs. For example, the smart contract to transfer transactions from a buyer is automatically executed upon receiving 132 a car from a seller and it does not require any third parties as an intermediary. In this architecture, the bullying identification   when health data or money is required to transfer from one account to another. A transaction typical contains sender, receiver address, digital signature, and health data. Digital signature and sender or receiver address are formed using PKI(Public Key 142 Infrastructure) that features anonymous properties of the transaction. A data transaction format is illustrated in Table I.  a higher stake has the highest probability to mine the next Block. But, PoS is less decentralized than that of PoW. The rich 158 node always mines the Block and becomes richer. We propose to divide Fog based Blockchain network into a different zone 159 as depicted in figure 3. The miner from a zone is nominated based on the following rules.

160
• The first round, let one node with the highest stake from a zone mine the next Block.

161
• The second round, let one node with the highest availability from a zone mine the next Block.

162
• The third round, let one node with higher processing capabilities from a zone mine the next Block.

163
• The fourth round, let one node with the smallest stake from a zone mine the next Block.

164
• The fifth round, let one node with a medium stake from a zone mine the next Block.

165
• The sixth round lets randomly one node from a zone mine the next Block.

166
The miner will be selected from each zone repeatedly according to the above-mentioned rule. More rules can be set to make 167 the selection more decentralized.

169
We utilized the Fog Gateway to prevent and detect the cyberbullying in a social network as a mediator. In a server-client 170 paradigm, a user has to directly experience the negative impact of cyberbullying if the bullying can not be prevented on time.  Finally, two outcomes produced by NLP and ML(Machine Learning) determines if the request will be committed or blocked.

224
If the NLP score is above a threshold value and ML generates "Yes", then the system blocks the stranger. If the NLP score is 225 below the threshold and ML generates "Yes", the comments of the stranger are committed for the first time but the stranger is 226 listed in the suspected list for fuhrer examining. If the NLP score is above the threshold but ML produces "No", the comment 227 of the stranger is sent to the user for the approval. If the NLP score is below threshold and ML produces "No", the comment 228 is committed without generating any warnings. The paper focuses on detecting cyberbully efficiently thus to give timely prevention. To achieve that, we utilize Blockchain 231 leveraged Fog based architecture with standard machine learning method and natural language processing technique. The 232 comparison of the proposed bullying detection framework with other existing bullying detection methods is presented in Table   233 II.   "fellate", "fellatio", "fingerfuck", "sex" ,"shag", "shagging", "shemale", "shit", "shitdick", "shite", "shithead shits shitted")  Set Waring to sender Address & discard to send message if there have any cyberbullying related information. Else send message to recipient address. 1. Select random_node from n fog nodes as miners 2. decoded_message = message_decoder (private_key, encoded_message) 3. Go to step 4 to 6 for cyberbullying detection 4. Get all cyberbullying related keyword as black_list from cloud database 5. Apply machine learning model on decoded_message to detect cyberbullying based on black_list 6. If(isCyberbullying) then execute line 7 to 9: 7.