Comprehensive Framework for Implementing Blockchain-enabled Federated Learning and Full Homomorphic Encryption for Chatbot security System

Comprehensive Framework for Implementing Blockchain-enabled Federated Learning and Full Homomorphic Encryption for Chatbot security System | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Comprehensive Framework for Implementing Blockchain-enabled Federated Learning and Full Homomorphic Encryption for Chatbot security System Nasir Ahmad Jalali, HongSong Chen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3862540/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 13 You are reading this latest preprint version Abstract Chatbot is an artificial intelligence application that can provide a conversational environment between human and machine. Most organizations and industries are willing to lay out their services through chatbot because it can provide 24/7 customer support. Meanwhile it raises security and privacy challenges like access control, data leakage during transmission, SQL injection attack, language model attack which make the users concerned about their data, performance and accuracy. Therefore this research paper proposed a comprehensive framework integrated blockchain, federated learning and fully homomorphic encryption algorithm with face recognition to solve above mentioned chatbot’s challenges. The experimental result shows that distributed system improves chatbot accuracy (90%) and more transaction in less time with more clients do not affect the performance. In contrast, more iteration and clients will decrease the accuracy, performance and transactions in centralized system. In addition, fully homomorphic encryption improve and speed up data encryption process. It encrypted more data (1792 MB) in a small amount of 1240 time/sec and conversation/transactions can be transferred via a secure network to ensure the confidentiality, integrity and authenticity of users’ data. The implementation of such comprehensive framework in real-life can improve chatbot security that is actively work as a customer agent in organization. Chatbot Security/Privacy Federated learning Full Homomorphic Encryption Blockchain Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Figure 15 Figure 16 Figure 17 Figure 18 Figure 19 Figure 20 1. INTRODUCTION Technological advancement with the involvement of businesses has led the world to develop chatbots to help businesses and industries provide assistance to their end users. [1]. Chatbot is composed of the two words chat and robot which is an automated computer program that can mimic human communication patterns to interact with humans via text and voice conversation (conversation of humans with machines). Chatbots are computer-related artificial programs that are developed in a conversational manner through various communication channels such as text messaging, voice, and mobile applications. Chatbots operate as virtual assistants that must play various roles and receive natural language input to generate intelligent responses and services, access data, and help users complete a specific task [2]. Chatbots try to understand users’ requests to provide a specific answer without human interference. Whenever the users’ query or conversation intersect with the chatbot's current knowledge, then the conversation will be passed to the human operator to provide the answer. Recently, modern chatbots can learn through machine learning algorithms during conversation with users [3]. People feel less comfortable talking to chatbots than talking to humans because they have less experience in dealing with chatbots, some of them even drop out of the conversation if they feel they are not talking to a real person. However, studies have shown that modern chatbots are based on conversation data associated with multiple data sources, making the conversation more natural than if they were talking to a real person [4]. In general, chatbots are just input/output computer programs, as shown in Fig. 1, pre-programmed by natural language processing (NLP) to be used instead of humans for classical conversations such as health advice, e-banking, and e-shopping, …etc., to avoid wasting time. On the other hand, people are cautious when it comes to their personal data because chatbots collect, learn and deal with users’ personal data, so this kind of chatbots leads to security challenges. Users do not know how to store, use, share, and handle their sensitive personal identity information (PII). The main goal of this paper is to investigate and discuss security, privacy, transparency, and data protection issues in order to find ways to minimize the above security issues through a systematic investigation in the era of chatbots to help organizations for improving the security of chatbots they are using as agent to offer the services. Figure 1 chatbot general system [5]. Financial chatbot is also a kind of chatbots that the financial organization provide their services through a chatbot to the users. It can perform transactions and give financial advice to users through chat or voice. They need to improve the security and privacy of chatbots to prevent fraud in the financial system [5]. 1.1 Contribution Nowadays, artificial intelligence based chatbots play an important role in business and industry. They work and process based on natural language processing (NLP) to provide services and give a specific answer according to the user's request. Since chatbots communicate over an Internet connection and collect huge amount of data for conversations, so it raises data security and privacy challenges during user communication and exchange the data for services inquiry. Therefore we can list our contributions as follows: We developed a framework for chatbots security and privacy system included three affective stages: Secure Access login Data Security in Transmission Data security in backend of chatbots This framework has integrated blockchain, federated learning, and homomorphic encryption algorithm with facial recognition. Based on our knowledge this is the first comprehensive framework for chatbot security that consist of various security procedures Concentration on chatbots security system defects and how to enhance users’ trust as well as chatbots accuracy and performance in the future. 1.2 Paper Organization The paper is organized as follows: Section 2 explains related work to the topic. Section 3 briefly explains architecture and application of chatbots system. Section 4 briefly explains chatbot as a services Section 5 reviews the system security challenges. Section 6 explains data collection and management challenges in chatbots. Section7 explains chatbots computation privacy issues such as homomorphic encryption, secure multiparty computations, and federated learning. Section 8 addresses security and privacy challenges in chatbots systems and developed comprehensive framework to improve chatbot security. Section 9 explains chatbot versus chatGPT, and Section 10 shows the conclusion. 2. RELATED WORK Recently, chatbots have attracted the attention of professionals, especially researchers, who are studying chatbot activities, where the security and privacy of the information collected or exchanged between the user and the chatbot are most important. While there are several researchers who have attempted to study information security and privacy issues in chatbots but users are still concerned about the security and privacy, transparency, integrity, and confidentiality of information collected by chatbots. According to our knowledge, there is currently no work that tackles the security and privacy of access or login, data transfer, and data in a backend of the chatbot. Previous work has only analysed the security and privacy of data transmission, access control, or security in the rest of the system, but not the complete security and privacy triad such as access or login control, secure data transmission, and security in the backend of chatbots. The author of [6] studied the security and privacy issues in his work and recommended the authentication and encryption method to improve the security and privacy only in the transmission part between chatbot and user. The goal of the re-study of [7] is to develop chatbots that support various functionalities to provide restrictive security measures to protect users' personal data during data transmission with chatbots in healthcare scenarios. The research of [8] examined various aspects of chatbots, including security and privacy issues, and found that chatbots faced some security challenges such as threats, vulnerabilities, data tampering, and data theft. Therefore, they recommended some solutions such as two-factor authentication, e.g., username and password, and sending a confirmation email or message containing security questions to overcome the above challenges, but these solutions are not responsive for chatbot security. The authors [9] stated in their research article that chatbots collect a huge amount of data during a conversation with users and store it. Users are concerned about losing control of their data after conversing with the chatbot, third party access to personal information, and inappropriate use of sensitive data. The solution based on their survey paper recommended that all technical parts of the chatbot should be improved for security and protection of users’ sensitive data. Another author [10] studied web-based chatbots to increase sales and provide timely and punctual responses to their customers for gaining insight into customer behaviour but in terms of security, they did not receive a guarantee because they had vulnerabilities such as tracking cookies and sharing data with third parties. After completing their survey, the authors found that despite the many websites promising privacy and security to customers, users are unknowingly exposed to insufficient security guarantees by companies selling their services using chatbots. The author's goal [11] was to get an overview of the security issues related to ChatGPT, such as malicious text, private data disclosure, fraudulent services, and unethical content production, so the author presented an empirical study to investigate the effectiveness of ChatGPT’s content filtering. The author also mentions in his conclusion that professionals, researchers and policy makers should try to improve research on the complex security challenges of ChatGPT, and the possible mitigation is to push content filtering, tag data, scan the output, or use artificial intelligence (AI) to filter AI content. The work of [12] author emphasized the security attacks and vulnerabilities that are associated with the general working modules of chatbots listed in Table 1 . The structure of a chatbot involves four main modules. The first module is the client module, which enables users to interact with the chatbot. The second module is the network module, which is responsible for sending messages to the response generation module and the database module. The third module is the response generation module, which generates answers to the input messages initiated by the users. Finally, the fourth module is the database module, which saves all the records generated by the clients and chatbots. Table 2 summarizes the viewpoint of various researchers on chatbots and security issues with their solutions and compares them with our research scheme 3. ARCHITECTURE DESIGN AND APPLICATION OF CHATBOTS SYSTEM Artificial intelligence has developed and popular rapidly in the recent era of technology like chatbot system which is the most popular AI-based technology to support and provide information to end users. It is an intelligent agent that can mimic human communication to interact with users to provide them with services as well as a set of appropriate answers to their questions and queries [13]. Chatbots are important technological system that can support, enhance and promote individual learning experiences in education, business and industry or banking services for customers [14]. The system architecture of chatbots mostly depends on the domain in which they are deployed, but AI chatbots consist of various components, as shown in Fig. 2, such as user interface, node server, environment, question and answer system, service system/intelligent automation [15]. Table 1 Chatbots vulnerabilities and cyber attacks Chatbot module Type of attack Impacts Descriptions Countermeasures Client Module [17] [18] • Fake response • Access control attack • Data leakage and loss trust • Session hijack and information loss • Attackers can manipulate chatbots by providing them with false and misleading information, causing them to perform malicious actions. • Unauthorized access to chatbots can result in the disclosure of sensitive information. • Response filtering • Regular testing • Strong authentication • Input validation Network Module [19] [20] • Dos Attacks • MiTM Attack • Services interruption • Identity theft and data manipulation • The chatbot interaction is stopped by a denial-of-service attack that floods the server with requests. • The adversaries intercepts the communication between two parties and replaces it with malicious content. • Strong Encryption • Load balancing Response Module [21] [22] • Language model attack • Adversarial test simple • Adversarial responding feedback • PII theft and trust loss • Wrong information • Reconstruct ML model to perform malicious tasks • Attackers make malicious attempts to steal information. • The input messages are crafted to lead fake response • An attacker can replicate the response generation module’s content to execute malicious attack task without modifying the model parameters • Verification Language model • Hate speech detector (future update) • "Network interpretations and data transformation with Style Transformer." Database Module [23] • SQL injection • Knowledge graph attack • Data deletion, alteration or theft • Loss of accuracy • An attacker can submit a harmful SQL statement, which can result in gaining unauthorized access to sensitive data. • An attacker can manipulate the knowledge graph of a chatbot in order to provide incorrect information. • Input validation • Storage mechanism • Identification detection • Access control • Strong encryption Table 2 The comparison of prior work with our developed framework Chatbot Name Research study topic Part of Chatbot Limitation Our developed framework Deep learning mechanism [24] [25] Chatbots using Deep Learning Architecture • Take more time for training. • Costly techniques. • No Standard Framework. We developed a comprehensive blockchain enabled federated learning framework with the integration of fully homomorphic encryption and facial recognition algorithm that can improve privacy / security, accuracy, performance and data management or transparency during three different stages such as login access, data security in transmission, and data privacy/security at the backend of chatbots. Health Care chatbot [26] [6] Design of Chatbot using Depp Learning Architecture • Data wasn’t trained on time. • Slow processing • Lack of security framework Diagnostic chatbot [27] [28] Design and development of diagnostic chatbot for supporting primary healthcare system Application • Less accuracy the human doctor • Contain small database to store 150 diseases. • Weak of machine learning part PriBot [29] [30] PriBot: Conversation privacy with Chatbots Application/Security • Risk of malicious users to make the bot useless • The bot is closer to personal privacy assistant than to a smarter searching interface • Users less trust on PriBot about their data usage. Chatbots [31] [32] Chatbots: Security, Privacy, Data protection, and Social aspects Security/Privacy • Data manipulation on provider side. Speech review chatbot system [33] An reinforcement leaning-based speech censorship chatbot system Security/Privacy • The impact of data imbalance and attack detection and aggressive response detection considering multiple rounds of dialogue. Figure 2 shows that the user's request via message or voice is forwarded through interface to the Node Server to find an appropriate response. Here, the Node Server is in contact with different parts; the environment is the main location for the natural language process for context explanation and has several sub-components; 1 - NLP engine is the main part of the environment that interprets what the user says in conversation time and converts it into structured data for further processing; the intent classifier takes the user request, identifies its meaning, and tracks it back to one of the chatbot supporting intents, the entity extractor extracts the main information of the user request. The Dialog Management Agent manages the current context of the dialog between the user and the chatbot to learn more from the user feedback for user future satisfaction. Question and answer is the main part in the chatbot system that interprets users' frequently asked questions to give users an accurate answer based on its knowledge. This part can give the answer based on two common methods: Manual training, where the company provides a list of questions and answers (Q&A) for the chatbot to find the answer from this list and send it back to the user, and automated training, where the company provides all the related documents and then the chatbot trains through machine learning based on the documents and provides the answer to the user query [34]. 4. CHATBOTS AS A SERVICE Today’s connected world has changed the way of businesses and consumers communication that is commonly known as artificial intelligent applications called chatbots. It is one of the most important forms of communication that can manage the relationship between service providers and customers through a conversational interface such as Facebook, WhatsApp, WeChat, etc., as shown in Fig. 3 ; which is integrated using natural language processing [35]. In education, health, industry, banking, and government, chatbots are used to provide real-time communication services to their customers to improve communication and save time and cost because chatbots need to collect and store a large amount of data about the user and the associated conversation [36]. it shows that a user generates a request process through a massaging interface via a speech or text input application and sends it to a natural language parser to convert it into the programming language of the conversation learning engine, or the conversation engine analyses the user request and sends it to the backend of the system, which is connected to many different databases that provide the response to user requests. Most messaging platforms are supported by third-party chatbots and they collect and store a large amount of data related users. Also, they need permission to add more features to enhance the conversation to complete the task on behalf of the user, so the security and privacy of such huge data is a critical issue in the chatbot system, therefore users and businesses need to be aware of data sharing protection [37]. 5. CHATBOT SYSTEM SECURITY CHALLENGES We have explained the architecture of chatbots in the previous section that chatbots collect and store a large amount of data through communicating with users over the Internet to provide the services, as well as exchange a piece of specific data between user and chatbot system application during the communication. Since chatbots contact with people over the Internet, they are vulnerable to cybersecurity and malicious activities. Therefore, it is very important to investigate data security and confidentiality during transmission and the rest of the system to increase users' confidence in the future use of chatbots [38]. 5.1 Security and privacy challenges Information security is the customer's willingness to preserve and control their personal identity information (PII), so privacy protection consists of three stages. First, the collection of personal data, second, control over the data, and third, knowledge of laws and regulations related to data processing and privacy [39]. Chatbots are used in various application areas where a large amount of data needs to be processed. Therefore, security and privacy are critical issues for chatbot systems, especially for the organization that chatbot uses for critical tasks such as financial information, data analytics, etc. Here, we can highlight that secure communication and authentication, data integrity, confidentiality and system availability, transparency and accountability are security challenges for chatbot systems [40]. First of all, it should be ensured that only authenticated users can communicate with chatbots to query/transmit their sensitive information related to financial issue. Second, the integrity and confidentiality of the data must be ensured. The transmitted data can be viewed only by authorized participants and should be protected from any kind of violation and corruption. Third, availability, which protects the system from interruptions and keeps it available to users. Transparency and accountability increase the trustworthiness of the chatbot system. The privacy and security issues arise from the lack of comprehensive security framework to control over the data transferred between the user and the chatbot system [37]. 5.2 AI based chatbots Security threats With the development of AI technologies, chatbots are able to reproduce human aspects such as conversation through speech, and text with high accuracy like humans. This opens the door to malicious activities and gives a chance to social engineering, man-in-the-middle, and phishing attacks for compromising sensitive information like hackers use bots as tools to imitate humans to trick them into submitting their payment details [40]. The victim unknowingly trusts chatbot requests and does not know that the chatbot is controlled by a cybercriminal group, so hackers collect a large amount of personal data from users through chatbots from numerous conversations with end users every day [41]. 5.3 Chatbots vulnerabilities Vulnerabilities are the weaknesses of chatbots that are exploited by adversaries to compromise the security of the system. These vulnerabilities arise from inadequate code, carelessness in code and infrastructure, unprotected and pornographic human errors and make the chatbot system vulnerable to cyber-attacks. Many chatbots use cloud computing services, which have their own threats and vulnerabilities, to store and process the data. So, it is the responsibility of the bot manufacturer to ensure all security processes related to chatbots, who is responsible for restoring the architecture and data flow, which should be encrypted both in transit and at rest in the system environment [42]. 5.3.1 End-to-End Encryption issue in chatbot This phase performs two functions; first. Transmission of user request to responded agent; second. Transmission user request from the responded agent to the database for providing the information. Communication with chatbot is associated to upper layer of Open System Interconnection (OSI) model so we can concentrate on OSI upper layer (transport-application) [43]. The upper layer of OSI model also face security threats even in some cases the adversaries uses various tools to lunch attacks such as Man-in-the-Middle (MitM) intercepts the legal conversation and modifies it with his malicious message or accessing encrypted data to extract sensitive information [44]. Most of time denial of services attacks also via communication phase that adversary’s access to legal conversation and alter it with their own information to participate in communication as legal client for communication disruption with server [45, 46]. 6. DATA COLLECTION AND MANAGEMENT PROBLEMS IN CHATBOTS Chatbots process a huge amount of data and store it in the system, whether the data is structured or unstructured, but data management is a major concern for users because they do not know what the chatbot does with a collected and stored data? How can data be shared without losing integrity and transparency? These are the two most common privacy questions that make users concerned about the security and protection of their data in the system [47]. Security techniques such as encryption, authentication, and verification can improve data protection of such large amounts of data. Therefore, data scientists have recommended and applied natural language processing (NLP) to address and overcome the security challenges in chatbots, because NLP is an artificial intelligence field that can analyse how computers interpret and manage speech and text to collect the information according to the concept of human language to provide a suitable mechanism for computer systems to manage human language and perform various related tasks [48]. Despite the advantages of natural language processing, it is also vulnerable to attack by test-time attackers. These vulnerabilities allow attackers to modify input to address model flaws. The universal attacker tries to find typically ungrammatical phrases to use as inputs for predicting bugs as well as for training-time attacks, such as Poisoning attacks where an attacker injects some malicious codes into the victim’s dataset as shown in Fig. 4 [49, 50]. The Fig. 4 explores that adversaries are inject a malicious code or application in dataset while it is training data model and make a backdoor for adversaries to do more malicious activities in the future for effecting users data and send them a triggered data instead of normal and corrected sentences [51]. 7. CHATBOTS COMPUTATION PRIVACY Chatbots are artificial intelligence-based computer programs that can process human conversations both spoken and written and allow people to interact with it as a real person. According to this nature of chatbot, it performs simple tasks, such as answering simple predefined questions and complicated tasks, such as digital assistants. Natural language processing performs all process regardless of chatbots type. On the other hand, computation is the critical part of chatbot and its privacy is very important to process the data and answer the users accurately. Therefore, the developers and owners of chatbots should pay attention to improve the protection of data processing by implementing security and privacy technologies and algorithms [52].We will briefly explain some of them; 7.1 Homomorphic Encryption The collection, handling and processing of sensitive data in chatbots or other AI-powered applications is very important and sensitive for both data and system owners, who must apply strict rules and processing algorithms. Homomorphic encryption is the appropriate and cost-effective solution to ensure confidentiality and prevent unauthorized access to personal and business data [53]. Homomorphic encryption is a collection of encryption algorithms that mimic and implement homomorphic properties, which perform a certain type of operations directly on the encrypted text and provide the same response as on the original message after decryption [54] as depicted in Fig. 5. In the field of cryptography, homomorphic encryption is a type of encryption scheme that allows third parties to perform computations on encrypted data. Therefore, encryption is an essential mechanism for maintaining the confidentiality of sensitive information, while conventional encryption mechanisms cannot perform this process on encrypted data [55]. 7.2 Secure Multi-Party computing This is a technique that allows multiple parties to perform computations together while keeping their input secret. Secure multiparty computation is the solution to various problems in joint computation without compromising data confidentiality, and ensures data confidentiality, independence, and accuracy for all parties involved in the computation [56] . Secure multiparty computation can perform inference on encrypted data, but is less suitable for training large language models such as chatbots because it includes a single dataset owned by a single entity and SMPC leads to increased computation and communication overhead, which has significant impact on system performance [57] . 7.3 Federated learning Artificial intelligence applications such as chatbots collect, store, and process massive amounts of data, which impacts performance, computing power and time, quality of service, and most importantly, privacy and security of user data. Due to privacy and security concerns, most data owners and users are not interested in sharing data with chatbots because chatbots are data-driven applications and collect user data from text, voice, or multimedia conversations [58].Privacy and security of user data during training of deep learning models may be violated by shared datasets. As mentioned earlier, there are many approaches to data security and privacy, but all of them require access to user data. In contrast, federated learning is an approach to improve data security and privacy where data is not collected in a central location, but the user’s data remains in its own location. Federated learning update the information through transferring deep leaning model which is trained over client’s private dataset as shown in Fig. 6 , this way federated learning protects data privacy and security [59, 60]. 7.4 Blockchain Blockchain technology employs a distributed database and ledger to securely timestamp and record transactions in a system. The primary idea behind blockchain technology is to provide a secure environment for anyone who communicate for exchanging the information through a public connection [61]. Blockchain operates on a peer-to-peer network where each node having a permanent and immutable copy of the ledger that keeps track of the entire map and the number of chained blocks using a hash technique [62, 63]. It is commonly used to securely transport or exchange information over a network. Unlike traditional centralized systems, which are controlled by a single entity, blockchain is a decentralized network of computers (nodes) where, each node keeps a copy of all transactions. If a new user or node wants to join and create the block of transactions to the system, then various consensus procedures like proof of work and proof of stake are used to validate and provide the agreement of network nodes (miners) about new adding node and block for transactions [64, 65]. Blockchain employs cryptographic techniques to safeguard transactions and regulate the generation of new units as depicted in Fig. 7. Public and private keys are used to secure transactions and restrict access to the blockchain. Once a block is uploaded to the blockchain, it is exceedingly impossible to change or delete the data contained within it. The blockchain’s immutability is achieved through cryptographic hashing and the consensus method, resulting in a tamper-resistant ledger [49] 8. ADDRESSING THE SECURITY AND PRIVACY CHALLENGES IN CHATBOTS SYSTEMS As mentioned earlier that chatbots are automated computer application which effected people live in various aspects like providing personal assistance, information, offering services and so on. Whenever a human-like dialog system has been developed known as chatbots the attention must be paid to potential security and privacy challenges, vulnerabilities that can lead data leaked and exploited. There are various type of security and privacy challenges that affect different parts of chatbot according to its architecture [67]. To better understand we have divided chatbots architecture into three parts where every part experience various security and privacy challenges with their own solutions; 8.1 Access control and Authentication The widespread uses of chatbots increases security and privacy risk especially for financial service providers that is harmful for both users and system. First layer to improve security/privacy of chatbot is to control login access to prevent illegal login to chatbot [31]. Access control can improve data integrity and confidentiality because it strongly restricts unauthorized users access to chatbots [68]. Therefore to have strong login access control we recommend a layer combined from username/password and facial recognition because it allow the chatbot to be connected with only with authorized and legal user who will also be accountable for his/her future actions [69]. 8.2 Data Transmission Security/Privacy Artificial intelligence based chatbots communicate with users and exchange data over a public internet connection. End-to-end encryption is a type of communication where only the legal and authorized parties (source and destination) can see, read and decrypt the message without anyone else. Therefore it is very important to secure the communication between user and chatbot to ensure that adversaries cannot access user data during transmission but only legal user can user encrypt and decrypt keys to read the messages [70]. We would like to use a new and secure algorithm (full Homomorphic Encryption) to improve the security of data during transmission because it allow the process to be performed over encrypted data in the contrast of other encryption methods and algorithms. Traditional encryption algorithms spread the concept of distributed keys where public and private keys are exchanging over internet during communication because they need process over decrypted data. In contrast, full homomorphic encryption algorithm (FHE) allows a complex mathematical operation to be performed on encrypted data during transmission without decryption [71]. FHE consist of three steps for data encryption process; key generation, encryption and decryption. Key generation step is randomized that take security parameters as input and generate public (PK) and secret key (SK) to encrypt data. Encryption step also randomized that takes PK and plaintext as input and generates the cihpertext, while decryption needs SK and ciphertext as input and generated plaintext [72], as depicted in Fig. 6 . Even though the adversaries access to data during transmission but they cannot decrypt and read the information therefore it reduce the risk of man-in-the-middle attack. Form another hand, full homomorphic encryption can support encrypted image processing, For example chatbot get the image from user and forward it to deep learning model for further processing and the result is transmitted back to chatbot to be displayed for user accordingly [73]. So the implementation of FHE can improve data security/privacy during transmission between user and chatbot [74]. 8.3 Data Storage and Security in Backend As earlier mentioned that chatbot collect a huge amount of data to a single point, so the processing of such huge data is challengeable and big issue. Chatbot are used in various application domain so it needs to keep data integrity, confidentiality, transparency, and accountability. From another hand, users also concerned about their data that how chatbot deal with it and how it share the data by keeping integrity and transparency [75]. To address these challenges and users concerns we developed a framework that integrated blockchain enabled federated learning (see section 8.4 ) to improve data security/privacy and transparency in the backend of chatbot. Blockchain is a distributed ledger technology that maintains a continuously growing ledger on the network to provide a secure transaction with a timestamp recording mechanism to support data security and privacy [76, 77]. Ledger is systematic system for data structure that consist of many blocks chained by a cryptographic mechanism. Therefore it uses a chained blocks to store and transmit data with public and private keys to verify and sign the transactions because it calls blockchain [78]. Ledger technology make the smart contract and its execution immutable, irreversible, and undesirable. In addition blockchain provide data persistence, distributed data control, data management, transparency and accountability. Blockchain is used and deployed in various fields to provide a secure environment for data processing. Recently, it can also be used with natural language processing because blockchain stores and share the data with other users through a distributed ledger [79]. Blockchain is used to store data as a smart contract in natural language processing to improve storage mechanism for trustworthiness and prevent SQL injection attack which are important for secure computation. In addition blockchain can store sentiment analysis data as smart contract in natural language processing to allow the users to access it as open source rather than closed or preparatory source that is cheap and easy for small businesses [80]. Blockchain helps the artificial intelligent system to process the data accurately and effectively by using smart contracts to connect different databases. Therefore, it makes data analysis true and the decision-making process becomes better for the organization. Therefore, Blockchain NLP allows users to compose a text through their voice or simply type the text using the keyboard, which is processed by artificial intelligent model to protect it from language model attacks and malicious injections [81, 82]. 8.4 Proposed Blockchain enabled fusion chatbots Security/Privacy framework This framework has been developed to improve chatbots security for facilitating some security functions to perform critical functionalities in a secure manner. The framework integrates blockchain-based federated learning and homomorphic encryption with face recognition that enhance the security/privacy of three correlative phases; (1)- Secure login access (2)- secure data transmission (3)- Data security, privacy and transparency in chatbots backend. Obviously the framework included two main functions front end for login access that user have direct access and backend for data storage and transactions but the third part is operating between frontend and backend for secure data transmission. Figure 9 shows the main framework that has two main parts front end and backend where the user has direct access to front end for performing the required options and backend stores transactions data. Figure 10 shows the flow diagram that describes the order and relationship between frond and backend parts and Fig. 11 shows the sequence diagram of main framework to explain the operation and interaction between users and chatbot. For framework implementation needs various datasets for chatbots such as NewQA dataset for questions and answer that contains over 10000 human generated questions and answers. This data set use Reading Comprehensive Model (RCM) to understand and interpret new articles and investigate new methods for handling complex questions and large documents. The Scheme Guided Dialog (SGD) dataset is used for dialog interactions. It consist of over 20k annotated multi-domain, task-oriented conversations between a human and virtual assistant. We would like to explain the structure of every part separately; 8.4.1 Research and evaluation method This research paper includes both theoretical and experimental method to evaluate some parts like the performance, accuracy and privacy of the proposed framework. First we performed a systematic research through various digital libraries study and carry out related research topics to shape a conceptual framework. Second, the developed framework tested in a chatbot virtual environment called RASA which is open source framework based on machine learning to create highly accurate text and voice based conversation chatbots. RASA action server provides the environment to write a code in python version 3.6 that mainly used to trigger actions of some parts like federated learning and FHE based on hyper ledger fabric with a related data sets mentioned in related parts. First. Secure login access: the user who wants to connect to chatbot he/she needs to be authenticated through first layer of security check which is included username/password and face recognition. Because it allows the chatbot to be connected only to a specific and authorized user who have access to sensitive data. He/she will also be accountable for their actions in the future, therefore it can improve data integrity and confidentiality by restricting unauthorized user access. The framwork used Viola-Jones algorithm for face detection with all features to train the classifier and we recommended convolutional neural network (CNN) which is a type of artificial neural network for picture classification, segmentation, processing and accuracy. CNN can extract the features from higher layers data to recognize image and it has the ability to develop internal representation of two dimensions of the image that allow the model to learn the positon and scale of the image. For the implementation it uses the Tufts face dataset which is a large-scale and public benchmark for face recognition. It is included 10000 image for different countries with age range form 4–74, the process are explained in Figs. 12, 13, and 14. Second. Secure Data Transmission: chatbots and user communicate through a public internet connection so it needs to be secure from unauthorized access to understand or alter the data, only the involved parties should be able to encrypt and decrypt the messages. Therefore we recommended a full homomorphic encryption algorithm for secure transmission because as we mentioned earlier. FHE allows the computation process to be performed on encrypted data without decryption. The computation process are explained in algorithm1. We tested it in a virtual environment including 5 computational clients were used to calculate the encryption time according to file size, which is performed on plaintext from 256–1792 bits data by adding 256 bits in each round. The Fig. 15 shows that fully homomorphic algorithm encrypt more data in less amount of time but its encryption time increased along with the size of plaintext. Figure 16 shows the fully homomorphic algorithm performance time for encryption and decryption according to keys size. Here the key sizes are selected as 512, 1024, 1536, 2048, 2560 bit to achieve 80,112,128,192,256 bit security levels. The encryption and decryption time has direct relationship to key size because of exponentiation operation that is increasing exponentially. Third. Data Security/privacy at backend: this is the big and important part of chatbots for data mining and processing where the user has access indirectly to ensure security and privacy as its flow diagram depicted in Fig. 10. We integrated blockchain enabled federated learning in framework for improving data security, privacy, transparency, accuracy and performance of chatbot. Federated learning is adequate mechanism to address such problem because federated learning use or exchange local and global model to update the information instead of collecting a huge amount of real data in a single central location. They only need to train local model over their personal dataset and send it to central server for aggregation to create global model. If we consider there are k clients over which the data has partitioned with \({P}_{k}\) as data set of indexes on \(k\) clients, so on each client \({n}_{k}=\left|{p}_{k}\right|.\) And local model training is conducting using local dataset as written in Eq. 1 $${w}_{t}^{k}={w}_{t-1}^{k}-\nabla wf\left({w}_{t-1}^{\left(c\right)}\right)$$ After local update it will be shared with server for aggregation, the process is shown in equations 2. $${F}_{k}\left(w\right)=\frac{1}{{}_{k}}\sum _{i={\rho }_{k}}fi\left(w\right)$$ $$gk=\nabla {F}_{k}\left({w}_{t}\right)$$ Where \({w}_{t}\) shows model weights in communication round \(t\) and \({w}_{t}^{k}\) indicates model weights on communication round \(t\) on client \(k\) , learning rate has shown by η, and \({p}_{k}\) shows set of data points on client \(k\) where \({}_{k}\) shows number of data points on client \(k\) . The Eq. 3 shows that central server aggregates all these gradients to apply the updates. $${w}_{t+1}\leftarrow {w}_{t}- \sum _{k=1}^{k}\frac{{n}_{k}}{n} gk$$ Collecting a huge amount of data to a single point will be risk full from the prospective of privacy and security and it can also affect performance and accuracy of computation. Figure 17 show the accuracy of federated learning (distributed) versus centralized system that FL has improved the accuracy of computation till 90 percent but the accuracy in centralized system is decreasing when the iterations get high (75 percent). Figure 18 shows the difference of performance between centralized and federated learning system that the federated learning (distributed) has high performance than centralized. Despite of this advantage of federated learning, there are still some concerns about data lose because federated learning operates centrally therefore the attackers can access to data trained models or can attack on central server for data leakage and access. Therefore we integrated blockchain technology to address the challenges of federated learning. Blockchain is more secure technology because it operates in decentralization manner. It stores all data on block-based manner, therefore once smart contract creates the block according to the consensus algorithm then it will not be changeable so it is more affective to improve data CIA triad. The system employs two smart contracts to manage user requests. The first contract, provided by the system, handles user registration and login. The second contract, on the other hand, is provided by the organization and is responsible for the encoding of the business logic for data transactions. For instance, in the case of a bank system, the smart contract simulates the functionalities of a financial institution. As such, every user of the system is expected to hold an account with the bank and perform financial activities such as balance inquiries and money transfers, which are carried out by the bank within the system. After conducting a thorough analysis of various public and private blockchains for deployment, we have concluded that public blockchains are more secure. However, they have some drawbacks such as being slow, open to all, and incurring significant costs to process and store data in a smart-contract like Ethereum. Therefore, we have decided to use a private blockchain system. Currently, Hyperledger Fabric is the most stable and popular private blockchain platform that supports smart contract functionality. This platform offers a unique concept of a channel that enables multiple blockchains to be managed within a single network. This creates a layer of confidentiality between various organizations to ensure that different activities remain private between different entities. Hyperledger Fabric uses several network entities like node, creator and validator showed in Fig. 9 . A smart contract in Fabric is referred to as chain and can be invoked through transactions. Node submitting a transactions to validator which is responsible to check the validation of an entity for its permission to perform an activity in a ledger encoded during transaction. The validation transactions status forwarded to a creator to create a new block according to transaction and return it to validation and node entity to update the ledger in blockchian as Fig. 19 Shows that transaction process has improved with high speed in our scheme from the prospective of user’s creation and transaction. The developed framework also includes sentiment analysis beside of technical and security parts that can evaluate customer interactions with chatbots. We used a lexicon based sentiment analysis techniques to extract the emotional polarity from the chat or text or users have with chatbots [83]. It examines all the words and sounds used by the customer during the interaction with the chatbot to understand the user's state, whether he is happy, sad, angry, or we can divide all these status on three major categories such as positive, negative and neutral status [84], Fig. 20 shows the flow diagram and algorithm2 the steps of sentiment analysis algorithm. Therefore, sentiment analysis can identify the areas that need to be improved. On the other hand, sentiment analysis is a form of cybersecurity perspective that refers to natural language processing and AI-based techniques to analyse users' state, attitudes, and opinions expressed in text and speech related to cybersecurity issues. Gathering information in the form of sentiment analysis that provides insight into how users think about services, specific products, or security issues is therefore key to improving the quality of services and security [85]. Algorithm 2: Sentiment Analysis Algorithm INPUT : Text chat ᴛ, the sentiment lexicon Ⅼ. OUTPUT : S mt = {P, Ng, or N} and Straight S, where P: Positive Ng: Negative, N:Neutral INITIALIZATION : SumPos and SumNeg = 0, where SumPos: accumulates the polarity of positive tokens t i−smt in T SumNeg: accumulates the polarity of negative tokens t i−smt in T Begin 1. For each t i ∈ T do 2. Search for t i in L 3. If ti ∈ Pos-list then 4. SumPos SumPos + t i-smt 5. Else if ti ∈ Neg-list then 6. SumNeg SumNeg + t i-smt 7. End if 8. End for 9. If SumPos > |SumNeg| then 10. S mt = P 11. S = SumPos/ (SumPos + SumNeg) 12. Else if SumPos < |SumNeg| then 13. Smt = Neg 14. S = SumNeg / (SumPos + SumNeg) 15. Else 16. Smt = N 17. S = SumPos/ (SumPos + SumNeg) 18. End If End 9. CHATBOTS VS CHATGPT Chatbots are an integral part of the digital world and are revolutionizing the way businesses interact with their customers. Therefore, the technological capabilities of chatbots are improving daily, from rule-based to complex conversational agents driven by artificial intelligence and machine learning algorithms. Although chatbots and ChatGPT are both conversational artificial intelligence technologies that have become increasingly popular in recent years, rises some security challenges and both are conversational agents that communicate with humans through natural language processing techniques [86, 87]. However, there are some minor differences between them in terms of structure, obtaining information, and generating responses to user queries as depicted in Table 3 . Chatbots are computer program which is designed to mimic human conversation through text or voice and are an application of artificial intelligence that creates an environment for communication between humans and machines in a conversational manner [32]. Chatbots focus on a specific domain, so they learn from decision trees or data predefined by the owner, as well as from user interactions. AI chatbots use natural language understanding and processing to generate human-like conversations. ChatGPT is a generative, pre-trained language model for chatbots developed by openAI. ChatGPT can handle a wide range of topics and domains because it uses deep leaning and a transform architecture to train the model through a vast amount of textual data on the web, enabling it to understand and generate human-like conversations [88]. 10. CONCLUSION In this paper, we conducted a survey research about chatbots security and privacy challenges. Some security threats, vulnerabilities and challenges existed to compromise chatbot privacy. Chatbot security and privacy is compromising via three stages: first. Login access, data during transmission and data at the backend, because chatbot always in contact with users to exchange and it collects a huge amount of data to be processed. As we analysed and investigated chatbot security system, the owners of chatbots only concentrated on communication and service providing aspect rather than security and privacy. Although some of the chatbots have security measures but they used very low level techniques that hackers can exploit it easily to access user’s data. Hence we developed a comprehensive and fusion framework to improve the privacy, security, accuracy, performance and encryption process of chatbot in all above mentioned security stages Table 3 Chatbot comparison with ChatGPT Attributes AI-Chatbots ChatGPT Architecture and design Machine learning model Generative Pre-trained Transformer Flexibility Flexibility based on predefined rule High flexibility Training Trained on specialized dataset Pre-trained on vast internet based data Conversational Depth Offer depth based on training data Offer more depth Personalization Can make personalized suggestions Personalization is extended Learning capabilities Learning from specific training data Learn from vast amount training data Use cases Task automation, customer support and information retrieval Creative writing tools, virtual assistance, chat experience Security and privacy Vulnerable to data breaches − User data collection − Data leak − Sharing confidential data − Algorithmic bias Vulnerable to data breaches − spread malicious software − business email compromise − user data collection − sharing confidential data − phishing emails during communication. This framework is integrated three fusion system: first. Face recognition system beside of normal authentication method (username and password) for secure and authorized login to prevent attacks related to login access. Second; A full homomorphic encryption algorithm for encrypted and secure transmission between user and chatbot to prevent malicious attacks like DoS and MiTM and access during transmission. Third. Blockchain enabled federated learning for data privacy, security and transparency at the backend of chatbots. The experimental test showed, the implementation of this framework improved accuracy and performance as well as prevent unauthorized users and attackers to exploit chatbot security while exchanging data during all above mentioned stages and it ensure the system that information cannot be shared with others. All conversations are transferred via a secure network and transactions are signed digitally by user private key to ensure the confidentiality of data and transaction without validated private key signature will not be processed, In addition, blockchain enabled FL is a distributed platform and protect the system against DoS and MiTM attacks. So we can summarize achievement of contributions in Table 4 : Table 4 summary table of contributions ( section 1.1 ) References Objectives Solutions Secure Login Access To improve login access in a chatbot we developed a framework that is included username/password and facial recognition because first time a user should register his/her face with chatbots and store the relevant information in a blockchain secure technology that is hard to be altered or tracked by unauthorized users. In addition we used CNN algorithm to process and classify all features related to picture for accuracy. Therefore facial recognition enable the chatbot to be connected with authorized user and restrict all illegal login access. Secure Data Transmission To have secure data transmission we have used fully homomorphic encryption algorithm because it allows the process to be performed faster over encrypted data in the contrast of other encryption algorithms. Only the source and destination are known about encryption and decryption key for data reading. Therefore adversaries can’t access data during transmission. Data Security/Privacy and Transparency in Backend of Chatbot To improve data privacy and trust on using chatbot we used federated learning where users only need to exchange trained model with central location instead of transferring the primary data for further process and chatbots functionalities. In addition we integrated blokchain with federated learning system in backend of chatbot to be responsive to the concerns of data transparency, security, confidentiality, performance and sharing with other users or parties. Because blockchain store the data in a block based and distributed manner which is include hash functions of previous block, therefore once the data transacted and stored in blockchain then no one can change because all participated nodes has the transaction copy. 11. FUTURE DIRECTION For future direction we recommend to enable chatbot especially financial chatbot based on beyond 5 generation mobile technology which is envisioned to revolutionize chatbots for delivering different services in the future. In addition I recommend to extend various chatbots such as social-bots testing against other vulnerabilities. Declarations Data availability all data sets analysed during this study to support the results of the article are publicly available. Author Contributions: authors contributed equally to the conceptualization and design of the solution for mentioned challenges. Data collection and analysis performed by Nasir Ahmad Jalali and Professor Chen Hongsong provide supervision as well as reviewed the paper for quality improvement. Acknowledgment This work was supported by the National Key Research and Development Program of China (No.2023YFC3303803 and 2023YFC3303800), State Key Laboratory of Public Big Data of Guizhou University (No.PBD2023-24), CCF NSFocus Kunpeng Foundation (No.CCF-NSFocus2023012) ，Fundamental Research Funds for the Central Universities (No. FRF-AT-19-009Z and FRF-AT-20-11) from the Ministry of Education of China. The authors declare that they have no financial or non-financial interests to disclose or personal relationships that could have appeared to influence the work reported in this paper. References C. B. N. &. C. A. B. Asbjorn Folstad, “What Makes Users Trust a Chatbot for Customer Services? An Exploratory Interview Study,” in The proceeding fof the Fifth International conferenceon Internet Science , Oslo, Norway, 2018. S. D. S. M. A. B. Lorenz Cuno Klopfenstein, “The Rise of Bots: A Survey of Conversational Interfaces, patterns, and Paradiams,” in DIS17: Proceedings of the 2017 Conference on Designing Interactiv System , 2017. P. Gentsch, “Conversational AI: How (Chat)Bots Will Reshape the Digital Experience,” in AI in Marketing Sales and Services: How Marketers without a Data Science Degree can user AI, Big Data and Bots , Frankfurt, Germany, SpringerLink, 2019, pp. 81-95. M.-T. L. D. R. S. e. a. Daniel Adiwardana, “Towards a Human-like Open-Domain Chatbot,” arXiv:2001.09977v3 [cs.CL] 27 Feb 2020, 2020. a. S. S. Takuma Okuda, “AI-Based Chatbot Service for Financial Industry,” FUJITSU Scientific and Technical Journal , vol. 54, no. 2, pp. 4-8, 2018. m. P. Milind H Shah, “Theoretical Evaluation of Securing Modules for Educational Chatbot,” in Proceedings of the Sixth International Conference on Intelligent Computing and Control System (ICICCS) , 2022. N. A. S. A. a. K. A. Mohammad Amin Kuhail, “Interacting with educational chatbots: A systematic review,” Education and Information Technologies, vol. 28, pp. 973-1018, 2023. h. A. J. P. a. V. S. Asim Mohammed Eltahir, “Review of Chatbot Security Systems,” in 2022 26th International Conference on Circuits, Systems, Communications and Computers (CSCC) , 2022. J. R. N. a. D. H. Rahime Belen Saglam, “Privacy Concerns in Chatbot Interactions: When to Trust and When to Worry,” in 23rd HCI International Conference, HCII 2021 , Swithzerland, 2021. M. I. S. S. H. e. a. Nazar Waheed, “An Empirical Assessment of Security and Privacy Risks of Web-Based Chatbots,” WISE international Journal, pp. 1-25, 2022. a. K. B. Erik Derner, “Beyond the Safguards: Exploring the Security Risks of ChatGPT,” arXiv:2305.08005v1 , 2023. M. B. Hoy, “Alexa, Siri, Cartana, and More: An Introduction to Voice Assistants,” Medical Reference Services Quarterly , vol. 37, no. 1, pp. 81-88, 2018. F. C. M. L. e. a. Fabio Clarizia, “Chatbot: An Education Support System For Students,” in International Symposium on Cyberspace Safety and Security , 2018. &. A. A.-I. Chinedu Wilfred Okonkwo, “Chatbots Applications in Education: A Systematic Review,” Comptuers and Education: Artificial Intelligence, vol. 2, 2021. X. Huang, “Chatbot: Design, Architecture, and Applications,” Univesity of Pennsylvania: School of Engineering and Applied Science , Pennsylvania, 2021. Dasagrandhi, “Understanding The Conversational Chatbot Architecture,” V-Soft Consulting, 2550 Eastpoint Pkwy Suite 300 Luisville, Ky 40223, 2020. D. J. M. L. e. a. Jing Xu, “Bot-Adversarial Dialogue for Safe Conversational Agents,” in Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , New York, 2021. N. M. e. a. Domna Bilika, “Hello Me, Meet the Real Me: Audio Deepfake Attacks on Voice Assistants,” arXiv:2302.10328v1 [cs.CR], 2023. M. L. J. V. a. S. L. Hyunji Chung, “Alexa, Can I Trust You?,” IEEE Xplore, vol. 9, pp. 100-105, 2017. S. B. a. P. Z. Krishna Gondaliya, “SLA as Mechanism to Manage Risks Related to Chatbot Services,” in 2020 IEEE 6th Intl Conference on Big Data Security on Cloud (BigDataSecurity) , 2020. H. Q. G. W. e. a. Cheng Qian, “Distinguish Sens from Nonsens:Out-of-Scope Detection for Virtual Assistants,” arXiv:2301.06544v1 [cs.CL] 16 Jan 2023, 2023. X. F. Z. X. e. a. Yang Zheng, “Why Adversarial Reprogramming works, when it fails, and how to tell the difference,” Information Sciences, vol. 632, pp. 130-143, 2023. F. S. a. C. T. Stefano Bistarelli, “A Chatbot Extended with Argumentation,” in 5th workshop on Advances In Argumentation In Artificial Intelligence , Milano, Italy, 2020. A. D. V. K. Neeraj Singh Kadayan, “Chatbot Using Deep Learning,” Journal of Emerging Technologies and Innovative Research (JETIR), vol. 6, no. 2, pp. 137-141, 2019. S. Z. E. e. a. Hana Emma Wube, “Text-Based Chatbot in Financial Sector: A Systematic Literature Review,” Data Science in Finance and Economics , vol. 2, no. 3, pp. 209-236, 2022. S. S. P. N. G. P. V. Nivila A, “Design of Chatbots Using Deep learning,” International Research Journal of Engineering and Technology (IRJET) , vol. 9, no. 20, pp. 1105-1110, 2022. N. R. Bushra Kidwai, “Design and Development of Diagnostic Chatbot for supporting Primary Health Care System,” Procedia Computer Science , vol. 167, pp. 75-84, 2020. a. S. D. Aishwarya Surani, “Understanding Privacy and Security Postures of Healthcare Chatbots,” Association for Computing Machinery (ACM), 2022. K. F. K. G. S. K. A. Hamza Harkous, “PriBots: Conversational Privacy with Chatbots,” in Workshop on the Future of Privacy Indicators, at the Twelfth Symposium on Usable Privacy and Security (SOUPS) , Denver, Colorado , 2016. A. R. M. S. F. e. a. Md. Saiful Islam Bhuiyan, “BONIK: A Blockchain Empowered Chatbot for Financial Transactions,” in Conference on Trust, Security and Privacy in Computing and Communications , 2020. J. N. e. a. Martin Hasal, “Chatbots: Security, Privacy, Data protection, and Social aspects,” Wiley, pp. 1-13, 2021. L. F. a. W. I. Minha Lee, “Brokerbot: A Cryptocurrency Chatbot in the Social-technical Gap of Trust,” Computer Supported Cooperative Work (CSCW), vol. 30, pp. 79-117, 2021. D. H. D. L. Z. Z. &. N. C. Shaokang Cai, “An Reinforcement Learning-based Speach Censorship Chatbot System,” The Journal of Supercomputing , vol. 78, pp. 8751-8773, 2022. N. Haristiani, “Artificial Intelligence (AI) Chatbots as Language Learning Medium: An inquiry,” in International Conference on Education, Science and Technology 2019 , Gothenburg, Sweden, 2019. L. B.-G. a. F. G. Jorge Cordero, “Use of Chatbots for Customer Serivces in MSMEs,” Applied Computing and Informatics , vol. 16, no. 1, 2022. a. S. H. Darius Zumstein, “Chatbots-and Interactive Technology for Personalized Communication, Transactions and Services,” IADIS International Journal on WWW/Internet, vol. 15, no. 1, pp. 96-109, 2018. C. M. a. F. P. Jide Edu, “Exploring the Security and Privacy Risk of Chatbots in Messaging Services,” in Proceeding of the 22nd ACM Internet Measurement Conference (IMC'22) , France, 2022. F.-Y. L. a. J.-W. L. Sen-Tarng Lai, “A Banking Chatbot Security Control Procedure for Protecting User Data Security and Privacy,” in 13th International Conference on Broadband and Wireless Computing Communication and Applications (BWCCA-2018) , Switzerland, 2019. L. P. &. A. D. V. Lucrezia Maria de Cosmo, “The role of attitude toward chatbots and privacy concern on the relationship between attitude toward mobile advertising and behavioral intent to use chatbots,” Italian Jouranal of Marketing, pp. 83-102, 2021. &. F. W. Josip Bozic, “Security Testing for Chatbots,” in In IFIP interanation Conference on Testing Software and System (ICTSS) , Switzerland, 2018. C. M. F. P. e. a. Jide Edu, “Exploring the Security and Privacy Risks of Chatbots in Messaging Services,” in In Proceedings of the 22nd ACM Internet Measurement Conference (IMC '22) , New York, USA, 2022. M. A. a. G. M. Koien, “Cyber Security and the Internet of Things: Vulnerabilities, Threats, Intruders and Attacks,” Journal of Cyber Security, vol. 4, pp. 65-88, 2015. a. R. H. Rohan Kar, “Applying Chatbots to the Internet of Things: Opportunities and Architectureal Elements,” International Journal of Advanced Comptuer Science and Application (IJACSA), vol. 7, no. 11, 2018. A. A. e. a. Avijit Mallik, “Man-in-the-middle Attack: Understanding in Simple words,” International Jouranl of Data and Network Science, vol. 3, pp. 77-92, 2019. D. S. e. a. Laura Feinstein, “Statistical Approaches to DDoS Attack Detections and Response,” in Proceedings of the DARPA Information Survivability Conference and Exposition , 2013. K. M. Y. N. a. Yameen Ajani, “Homomorphic encryption for Secure Conversation with AI bots over cloud to prevent Social Engineering attacks,” International Journal of Engineering Research and Applications, pp. 21-27, 2021. T. V. Pathrabe, “Survey on Security Issues of Growing Technology: Big Data,” in National Conference on Latest Trends in Networking and Cyber Security , 2017. S. Jung, “Semantic Vector Learning for Natural Language Understanding,” Science Direct, vol. 56, pp. 130-145, 2019. S. F. e. a. Eric Wallace, “Universal Adversarial Triggers for Attacking and Analyzing NLP,” in Proceedings of the 2019 Conferene on Empirical Methods in natural Language Processing , Hong Kong, China, 2019. B. W. R. G. e. a. Jiwei Tian, “Advarsarial Attacks and Defenses for Deep-Learning-Based Unmanned Aerial Vehicles,” IEEE Internet of Things Journal, vol. 9, no. 22, pp. 22399-22409, 2022. B. W. ,. Z. W. e. a. Jiwei Tiam, “Joint Adversarial Example and False Data Injections Attacks for State Estimation in Power Systems,” IEEE Transactions on Cybernetics, vol. 52, no. 12, pp. 13699-13713, 2022. T. K. D. a. C. P. Fran Casino, “A systematic literature review of blockchain-based applications: current status, classification and open issues,” Telematics and Informatics, vol. 36, pp. 55-81, 2019. Y.-L. C. L. Y. P. &. C. S. K. Jing Yang, “A systematic Literature Review of Information Security in Chatbots,” Applied Sciences, vol. 13, pp. 2-18, 2023. M. W. F. a. G. I. S. Michael Lahzi Gaid, “Secure Translation Using Fully Homomorphic Encryption and Sequence-to-Sequence Neural Network,” in 28th International Conference on Computer Theory and Application (ICCTA) , Alexandria, Egypt, 2018. a. H. C. Nasir Ahmad Jalali, “Federated Learning Security and Privacy-Preserving Algorithm and Experimental Research Under Internet of Things Critical Infrastracture,” Tsinghua Science and Technology , vol. 29, no. 2, pp. 400-414, 2023. H. A. a. A. U. Abbas Acar, “A Survey on Homomorphic Encryption Scheme: Theory and Implementation,” ACM Computing Surveys, vol. 51, no. 4, pp. 79:1-79:35, 2018. D. Romanov, “Secure Multi-party computation for supply chain collaboration,” Delft University of Technology, 2021. G. Sebastian, “Privacy and Data Protection in ChatGPT and Other AI Chatbots: Strategies for Securing User Information,” Georgia Institiute of Technology, 2022. N. A. J. &. H. Chen, “Security Issues and Solutions in Federated Learning Under IoT Critical Infrastructure,” Wireless Personal Communication, vol. 129, pp. 475-500, 2023. H. M. B. A. e. a. Peter Kairouz, “Advanced and Open Problems in Federated Learning,” Arxive: Machine Learning, vol. 14, no. 1-2, 2021. W. Z. Q. L. e. a. Renhao Lu, “Adaptive Asynchronous Federated Learning,” Future Generation Computer Systems, vol. 152, pp. 193-206, 2023. &. Q. L. Winson Ye, “Chatbot Security and Privacy in the Age of Personal Assistants,” in 2020 IEEE/ACM Symoposium on Edge Computing (SEC) , 2020. I. I. M. A. S. a. A. O. Peter Voege, “Smart Chatbot for User Authentication,” Electronics, vol. 11, 2022. S. N. &. M. K. Aneesa M P, “Face Recognition using CNN: A Systematic Review,” International Journal of Engineering Research & Technology (IJERT), vol. 11, no. 6, pp. 182-185, 2022. R. E. Endeley, “End-to-End Encryption in Messaging Services and National Security-Case of WhatsApp Messanger,” Journal of Information Security, vol. 9, pp. 95-99, 2018. G. Gentry, “A Fully Homomorphic Encryption Scheme,” Stanford University , California, 2009. A. G. e. a. Marcelo Blatt, “Secure Large-scale Genome-wide Association Studies Using Homomorphic Encryption,” Proceedings of the National Academy of Sciences of the United State of America (PNAS), vol. 117, no. 21, pp. 11608-11613, 2020. S. ,. M. A. Jamal N. Bani Salameh, “Evaluation of Cloud Computing Platform for Image Processing Algorithm,” Journal of Engineering Science and Technology, vol. 14, no. 4, 2019. K. M. e. a. Yammen Ajani, “Homomorphic Encryption Technology for Securing Data in Cloud Computing: A Survey,” International Journal of Computer Application, vol. 160, no. 6, 2017. L. T. Y. &. J. H. P. Sushil Kumar Singh, “FusionFedBlock: Fusion of Blockchain and Federated Learning to preserve privacy in industry 5.0,” Inforamtion Fusion, vol. 90, pp. 233-240, 2023. G. B. G. F. e. a. Daniele Mazzei, “A Blockchain Tokenizer for Industrial IoT Trustless Application,” Future Generation Computer Systems, vol. 105, pp. 432-445, 2020. Y. Q. Y. X. a. L. G. Chenhao Xu, “Asynchronous Federated Learning on Heterogeneous Devices: A Survey,” Comptuer Science Review , vol. 50, 2023. F. C. &. M. O. A. Md Sadek Ferdous, “In Search of Self-Sovereign Identity leveraging Blockchain Technology,” IEEE Access, vol. 7, pp. 103059-103079, 2019. &. M. N. Abylay Satybaldy, “Reveiw of Technical for Privacy-Preserving Blockchain Systems,” in Proceeding of the 2nd ACM International Symposium on Blockchain and Secure Critical Infrastructure , Taipei, Taiwan, 2020. ,. H. C. e. a. Pranab Kumar Bharimalla, “A Blokchain and NLP Based Electroni Health Record System: Indian subcontinent Context,” Informatica , vol. 45, pp. 605-616, 2021. Y. Z. J. Y. X. L. H. L. Kun Shao, “The Triggers that open the NLP model backdoors are hidden in th advaresial Samples,” Comptuer & Security , vol. 118, 2022. D. G. &. T. K. Ramanpreet Kaur, “Artificial Intelligence for Cybersecurity: Literature Review and Future Research Direction,” Information Fusion , vol. 97, 2023. O. F. F. T. I. a. I. A. O. Femi Emmanuel Ayo, “Machine Learning Techniques for Hate Speech Classification of Twitter Data: State-of-art, Future Challenges and Research Direction,” Computer Science Review , vol. 38, 2020. H. Z. j. M. e. a. Junfei Yi, “A review on the Covid-19 Pandemic prevention and control system based on AI,” Engineering Application of Artificial Intelligence , vol. 114, 2022. E. U. U. R. L. R. e. a. Ogobuchi Daniel Okey, “Investigating ChatGPT and cybersecurity: A Perspectived on topic modeling and Sentiment analysis,” Comptuers & Security, vol. 135, 2023. C. Chaka, “Geerative AI Chatbots-ChatGPT versus YouChat versus Chatsonic: Use Case of Selected Areas of Applied English Language Studies,” International Journal of Learning,Teaching and Eduactioanl Research , vol. 22, no. 6, pp. 1-19, 2023. A. A. a. Y. J. Muna AI-Hawawreh, “Chatgpt for Cybersecurity: Practical Applications, Challenges and future Directions,” Cluster Computing , vol. 26, pp. 3421-3436, 2023. I. C. O. K. e. a. Jan Kocon, “ChatGPT : Jack of all trades, master of none,” Information Fusion, vol. 99, 2023. Additional Declarations No competing interests reported. Supplementary Files Authorsdetails.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 17 Mar, 2024 Reviews received at journal 03 Mar, 2024 Reviewers agreed at journal 03 Mar, 2024 Reviews received at journal 23 Jan, 2024 Reviewers agreed at journal 17 Jan, 2024 Reviewers agreed at journal 17 Jan, 2024 Reviewers agreed at journal 17 Jan, 2024 Reviewers agreed at journal 17 Jan, 2024 Reviewers agreed at journal 17 Jan, 2024 Reviewers invited by journal 17 Jan, 2024 Editor assigned by journal 16 Jan, 2024 Submission checks completed at journal 16 Jan, 2024 First submitted to journal 14 Jan, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3862540","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":267578801,"identity":"5e504725-b293-4212-8f42-99df8ba40cd1","order_by":0,"name":"Nasir Ahmad Jalali","email":"","orcid":"","institution":"University of Science and Technology Beijing","correspondingAuthor":false,"prefix":"","firstName":"Nasir","middleName":"Ahmad","lastName":"Jalali","suffix":""},{"id":267578802,"identity":"7f87c5b2-98a6-4028-8a28-fd3d57c46935","order_by":1,"name":"HongSong Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAz0lEQVRIiWNgGAWjYBAC/hkMDMwMDDZAJg8QsxGhReIGWEsaUDWxWgwiwFoOk6JFuvnY44I/5xPnz+89wPCh7DAD/+wGAlpkjqUbz2y7nbjhGF8C44xzhxkk7hwgoEUix0yatwGohY3HgJm37TBQJIGQlvxv0jx/ziXObwNq+UuUlogcNmketgOJDceAWhiJ0SJxIw3osLZk4w3H8hIO9pxL55G4QUAL/4zkZ0CH2cnObz578MGPMms5/hkEtKCAAwyQNDAKRsEoGAWjgFIAANzUP+VNm+6RAAAAAElFTkSuQmCC","orcid":"","institution":"University of Science and Technology Beijing","correspondingAuthor":true,"prefix":"","firstName":"HongSong","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2024-01-14 07:29:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3862540/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3862540/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":49743103,"identity":"4facec0f-0b14-492f-912e-67681483fba6","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":53059,"visible":true,"origin":"","legend":"\u003cp\u003echatbot general system [5].\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/60e5f0b0305fe46004d2518a.png"},{"id":49743722,"identity":"ef3eb76a-0307-4b2d-9abe-4f8cc3ba6c00","added_by":"auto","created_at":"2024-01-17 09:58:49","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":97279,"visible":true,"origin":"","legend":"\u003cp\u003eAI Based conversational chatbot architecture adapted from [16]\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/fd7b233bc69e7bc626c26399.png"},{"id":49744339,"identity":"e117cdf6-065e-4138-ad4b-511c49e02795","added_by":"auto","created_at":"2024-01-17 10:06:49","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":71670,"visible":true,"origin":"","legend":"\u003cp\u003eChatbot as a service general structure adopted \u0026nbsp;[36]\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/3305e7e031254ec6ce5e6d02.png"},{"id":49745133,"identity":"180ec855-8815-41fd-af31-6a4cc59388da","added_by":"auto","created_at":"2024-01-17 10:22:49","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":108322,"visible":true,"origin":"","legend":"\u003cp\u003eoverview of poisoning with backdoor attack model training in NLP [51]\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/7f009268d795561d5b6707eb.png"},{"id":49743102,"identity":"698c4dbc-53f8-4632-b509-1cdb8c7fa515","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":75193,"visible":true,"origin":"","legend":"\u003cp\u003ehomomorphic encryption general scheme [55]\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/a2681acc142eab753b325de8.png"},{"id":49744612,"identity":"f0bcf037-9067-4177-a4ae-f5e134c3aa2f","added_by":"auto","created_at":"2024-01-17 10:14:49","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":59625,"visible":true,"origin":"","legend":"\u003cp\u003eFederated learning system and training structure [60]\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/9971601e318e0f6ee26fba04.png"},{"id":49743108,"identity":"b7697b78-da13-4572-8f7a-06de8a618064","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":98066,"visible":true,"origin":"","legend":"\u003cp\u003eblockchain and data blocks structure \u0026nbsp;[52]\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/0aa3fba24c6d3aba53fbd248.png"},{"id":49744342,"identity":"749de122-0518-4a94-8df9-78e195bffb30","added_by":"auto","created_at":"2024-01-17 10:06:49","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":45475,"visible":true,"origin":"","legend":"\u003cp\u003eHomomorphic encryption operational structure [74]\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/6b75f33639e5c1a1a6ac52df.png"},{"id":49743114,"identity":"53d6d0d9-0ff4-4169-9147-7941a49f14ab","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":178155,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eMain framework to enhance chatbot security and privacy\u003c/em\u003e\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/07c7bf6b9f13f19a41ec1c3f.png"},{"id":49744344,"identity":"c9db1964-9d4a-4533-9d8e-9134b622d010","added_by":"auto","created_at":"2024-01-17 10:06:49","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":123579,"visible":true,"origin":"","legend":"\u003cp\u003eflow diagram of main framework\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/1ca66eb6320fc66adef3d9fa.png"},{"id":49743121,"identity":"326d6f49-d168-43b9-97a5-f2ce15f9408b","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":126734,"visible":true,"origin":"","legend":"\u003cp\u003eSequence diagram for the proposed framework\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/5427ca1a1cd620eca0bb4be6.png"},{"id":49743118,"identity":"b7d44a9b-b8a7-4d47-9c41-0a8bdd6ac698","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":61542,"visible":true,"origin":"","legend":"\u003cp\u003eface recognition block diagram\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/0ef655a808818e02be1e6330.png"},{"id":49743111,"identity":"bdafb0d0-d79d-4158-9e86-1d26da33b74c","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":68191,"visible":true,"origin":"","legend":"\u003cp\u003eCNN structure for image data processing and classification\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/8ba74bd13efc2a779c9fc694.png"},{"id":49743724,"identity":"9bdedeb7-d8b1-44fd-88c9-0459b0448269","added_by":"auto","created_at":"2024-01-17 09:58:49","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":88944,"visible":true,"origin":"","legend":"\u003cp\u003eUser with Face registration sequence diagram\u003c/p\u003e","description":"","filename":"14.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/d51dd42c77e13390f59df2e1.png"},{"id":49744615,"identity":"6b9976a0-56b9-44de-9d3e-ecc70f2a9918","added_by":"auto","created_at":"2024-01-17 10:14:49","extension":"png","order_by":15,"title":"Figure 15","display":"","copyAsset":false,"role":"figure","size":70005,"visible":true,"origin":"","legend":"\u003cp\u003eFully homomorphic encryption time/file size\u003c/p\u003e","description":"","filename":"15.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/705fbecde226c06bc61b0477.png"},{"id":49744614,"identity":"38331df4-8c4e-4148-b95f-cfee124ff992","added_by":"auto","created_at":"2024-01-17 10:14:49","extension":"png","order_by":16,"title":"Figure 16","display":"","copyAsset":false,"role":"figure","size":152879,"visible":true,"origin":"","legend":"\u003cp\u003eFHE running encryption/decryption time per key size\u003c/p\u003e","description":"","filename":"16.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/e9183dec72be8d3dfc2e0371.png"},{"id":49743113,"identity":"10b86960-1106-445f-907f-29998e4d52d3","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":17,"title":"Figure 17","display":"","copyAsset":false,"role":"figure","size":94364,"visible":true,"origin":"","legend":"\u003cp\u003eAccuracy between centralized and distributed (FL)\u003c/p\u003e","description":"","filename":"17.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/32daec0527fc68bcd3aee838.png"},{"id":49743725,"identity":"615a30fe-5b12-496e-94b0-3e5911a71716","added_by":"auto","created_at":"2024-01-17 09:58:49","extension":"png","order_by":18,"title":"Figure 18","display":"","copyAsset":false,"role":"figure","size":206421,"visible":true,"origin":"","legend":"\u003cp\u003eperformance between centralized and distributed (FL)\u003c/p\u003e","description":"","filename":"18.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/d36cd28216f2b11c95ab733f.png"},{"id":49743729,"identity":"374fb1b9-968b-411f-9c3c-1a3758185d88","added_by":"auto","created_at":"2024-01-17 09:58:49","extension":"png","order_by":19,"title":"Figure 19","display":"","copyAsset":false,"role":"figure","size":21270,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of transactions and user creation\u003c/p\u003e","description":"","filename":"19.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/1ace2d479604cb29f69db1d7.png"},{"id":49743122,"identity":"255cbd0a-b623-432b-9312-d0553ce5359a","added_by":"auto","created_at":"2024-01-17 09:50:49","extension":"png","order_by":20,"title":"Figure 20","display":"","copyAsset":false,"role":"figure","size":38341,"visible":true,"origin":"","legend":"\u003cp\u003eSentiment analysis flow diagram\u003c/p\u003e","description":"","filename":"20.png","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/46bed777bb3512dc5af03b5b.png"},{"id":49992732,"identity":"4226e0a4-7b43-45ac-880c-a327e9f46684","added_by":"auto","created_at":"2024-01-22 18:58:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2316330,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/c4d30ed2-f2cc-4d43-a12f-c30883d5949f.pdf"},{"id":49743718,"identity":"3c31b646-4191-4f53-9375-2415c6094f51","added_by":"auto","created_at":"2024-01-17 09:58:49","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":53509,"visible":true,"origin":"","legend":"","description":"","filename":"Authorsdetails.docx","url":"https://assets-eu.researchsquare.com/files/rs-3862540/v1/e4adecc323a366ec5402a63a.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Comprehensive Framework for Implementing Blockchain-enabled Federated Learning and Full Homomorphic Encryption for Chatbot security System","fulltext":[{"header":"1. INTRODUCTION ","content":"\u003cp\u003eTechnological advancement with the involvement of businesses has led the world to develop chatbots to help businesses and industries provide assistance to their end users. [1]. Chatbot is composed of the two words chat and robot which is an automated computer program that can mimic human communication patterns to interact with humans via text and voice conversation (conversation of humans with machines). Chatbots are computer-related artificial programs that are developed in a conversational manner through various communication channels such as text messaging, voice, and mobile applications. Chatbots operate as virtual assistants that must play various roles and receive natural language input to generate intelligent responses and services, access data, and help users complete a specific task [2].\u003c/p\u003e \u003cp\u003eChatbots try to understand users\u0026rsquo; requests to provide a specific answer without human interference. Whenever the users\u0026rsquo; query or conversation intersect with the chatbot's current knowledge, then the conversation will be passed to the human operator to provide the answer. Recently, modern chatbots can learn through machine learning algorithms during conversation with users [3]. People feel less comfortable talking to chatbots than talking to humans because they have less experience in dealing with chatbots, some of them even drop out of the conversation if they feel they are not talking to a real person. However, studies have shown that modern chatbots are based on conversation data associated with multiple data sources, making the conversation more natural than if they were talking to a real person [4].\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn general, chatbots are just input/output computer programs, as shown in Fig.\u0026nbsp;1, pre-programmed by natural language processing (NLP) to be used instead of humans for classical conversations such as health advice, e-banking, and e-shopping, \u0026hellip;etc., to avoid wasting time. On the other hand, people are cautious when it comes to their personal data because chatbots collect, learn and deal with users\u0026rsquo; personal data, so this kind of chatbots leads to security challenges. Users do not know how to store, use, share, and handle their sensitive personal identity information (PII). The main goal of this paper is to investigate and discuss security, privacy, transparency, and data protection issues in order to find ways to minimize the above security issues through a systematic investigation in the era of chatbots to help organizations for improving the security of chatbots they are using as agent to offer the services.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eFigure\u0026nbsp;1\u003c/strong\u003e \u003cp\u003e \u003cem\u003echatbot general system [5].\u003c/em\u003e \u003c/p\u003e \u003c/p\u003e \u003cp\u003eFinancial chatbot is also a kind of chatbots that the financial organization provide their services through a chatbot to the users. It can perform transactions and give financial advice to users through chat or voice. They need to improve the security and privacy of chatbots to prevent fraud in the financial system [5].\u003c/p\u003e \u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e1.1 Contribution\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eNowadays, artificial intelligence based chatbots play an important role in business and industry. They work and process based on natural language processing (NLP) to provide services and give a specific answer according to the user's request. Since chatbots communicate over an Internet connection and collect huge amount of data for conversations, so it raises data security and privacy challenges during user communication and exchange the data for services inquiry. Therefore we can list our contributions as follows:\u003c/p\u003e\u003cul\u003e \u003cli\u003e \u003cp\u003eWe developed a framework for chatbots security and privacy system included three affective stages:\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e\u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eSecure Access login\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eData Security in Transmission\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eData security in backend of chatbots\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eThis framework has integrated blockchain, federated learning, and homomorphic encryption algorithm with facial recognition.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eBased on our knowledge this is the first comprehensive framework for chatbot security that consist of various security procedures\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eConcentration on chatbots security system defects and how to enhance users\u0026rsquo; trust as well as chatbots accuracy and performance in the future.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e\u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e1.2 Paper Organization\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eThe paper is organized as follows: Section 2 explains related work to the topic. Section 3 briefly explains architecture and application of chatbots system. Section 4 briefly explains chatbot as a services Section 5 reviews the system security challenges. Section 6 explains data collection and management challenges in chatbots. Section7 explains chatbots computation privacy issues such as homomorphic encryption, secure multiparty computations, and federated learning. Section 8 addresses security and privacy challenges in chatbots systems and developed comprehensive framework to improve chatbot security. Section \u003cspan refid=\"Sec24\" class=\"InternalRef\"\u003e9\u003c/span\u003e explains chatbot versus chatGPT, and Section \u003cspan refid=\"Sec25\" class=\"InternalRef\"\u003e10\u003c/span\u003e shows the conclusion.\u003c/p\u003e \u003c/div\u003e"},{"header":"2.\tRELATED WORK","content":"\u003cp\u003eRecently, chatbots have attracted the attention of professionals, especially researchers, who are studying chatbot activities, where the security and privacy of the information collected or exchanged between the user and the chatbot are most important. While there are several researchers who have attempted to study information security and privacy issues in chatbots but users are still concerned about the security and privacy, transparency, integrity, and confidentiality of information collected by chatbots. According to our knowledge, there is currently no work that tackles the security and privacy of access or login, data transfer, and data in a backend of the chatbot. Previous work has only analysed the security and privacy of data transmission, access control, or security in the rest of the system, but not the complete security and privacy triad such as access or login control, secure data transmission, and security in the backend of chatbots.\u003c/p\u003e \u003cp\u003eThe author of [6] studied the security and privacy issues in his work and recommended the authentication and encryption method to improve the security and privacy only in the transmission part between chatbot and user. The goal of the re-study of [7] is to develop chatbots that support various functionalities to provide restrictive security measures to protect users' personal data during data transmission with chatbots in healthcare scenarios. The research of [8] examined various aspects of chatbots, including security and privacy issues, and found that chatbots faced some security challenges such as threats, vulnerabilities, data tampering, and data theft. Therefore, they recommended some solutions such as two-factor authentication, e.g., username and password, and sending a confirmation email or message containing security questions to overcome the above challenges, but these solutions are not responsive for chatbot security. The authors [9] stated in their research article that chatbots collect a huge amount of data during a conversation with users and store it. Users are concerned about losing control of their data after conversing with the chatbot, third party access to personal information, and inappropriate use of sensitive data. The solution based on their survey paper recommended that all technical parts of the chatbot should be improved for security and protection of users\u0026rsquo; sensitive data. Another author [10] studied web-based chatbots to increase sales and provide timely and punctual responses to their customers for gaining insight into customer behaviour but in terms of security, they did not receive a guarantee because they had vulnerabilities such as tracking cookies and sharing data with third parties. After completing their survey, the authors found that despite the many websites promising privacy and security to customers, users are unknowingly exposed to insufficient security guarantees by companies selling their services using chatbots. The author's goal [11] was to get an overview of the security issues related to ChatGPT, such as malicious text, private data disclosure, fraudulent services, and unethical content production, so the author presented an empirical study to investigate the effectiveness of ChatGPT\u0026rsquo;s content filtering. The author also mentions in his conclusion that professionals, researchers and policy makers should try to improve research on the complex security challenges of ChatGPT, and the possible mitigation is to push content filtering, tag data, scan the output, or use artificial intelligence (AI) to filter AI content. The work of [12] author emphasized the security attacks and vulnerabilities that are associated with the general working modules of chatbots listed in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The structure of a chatbot involves four main modules. The first module is the client module, which enables users to interact with the chatbot. The second module is the network module, which is responsible for sending messages to the response generation module and the database module. The third module is the response generation module, which generates answers to the input messages initiated by the users. Finally, the fourth module is the database module, which saves all the records generated by the clients and chatbots. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e summarizes the viewpoint of various researchers on chatbots and security issues with their solutions and compares them with our research scheme\u003c/p\u003e"},{"header":"3.\tARCHITECTURE DESIGN AND APPLICATION OF CHATBOTS SYSTEM","content":"\u003cp\u003eArtificial intelligence has developed and popular rapidly in the recent era of technology like chatbot system which is the most popular AI-based technology to support and provide information to end users. It is an intelligent agent that can mimic human communication to interact with users to provide them with services as well as a set of appropriate answers to their questions and queries [13]. Chatbots are important technological system that can support, enhance and promote individual learning experiences in education, business and industry or banking services for customers [14].\u003c/p\u003e \u003cp\u003eThe system architecture of chatbots mostly depends on the domain in which they are deployed, but AI chatbots consist of various components, as shown in Fig.\u0026nbsp;2, such as user interface, node server, environment, question and answer system, service system/intelligent automation [15].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eChatbots vulnerabilities and cyber attacks\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChatbot module\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eType of attack\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eImpacts\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDescriptions\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCountermeasures\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClient Module [17] [18]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull; Fake response\u003c/p\u003e \u003cp\u003e\u0026bull; Access control attack\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull; Data leakage and loss trust\u003c/p\u003e \u003cp\u003e\u0026bull; Session hijack and information loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; Attackers can manipulate chatbots by providing them with false and misleading information, causing them to perform malicious actions.\u003c/p\u003e \u003cp\u003e\u0026bull; Unauthorized access to chatbots can result in the disclosure of sensitive information.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026bull; Response filtering\u003c/p\u003e \u003cp\u003e\u0026bull; Regular testing\u003c/p\u003e \u003cp\u003e\u0026bull; Strong authentication\u003c/p\u003e \u003cp\u003e\u0026bull; Input validation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNetwork Module [19] [20]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull; Dos Attacks\u003c/p\u003e \u003cp\u003e\u0026bull; MiTM Attack\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull; Services interruption\u003c/p\u003e \u003cp\u003e\u0026bull; Identity theft and data manipulation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; The chatbot interaction is stopped by a denial-of-service attack that floods the server with requests.\u003c/p\u003e \u003cp\u003e\u0026bull; The adversaries intercepts the communication between two parties and replaces it with malicious content.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026bull; Strong Encryption\u003c/p\u003e \u003cp\u003e\u0026bull; Load balancing\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResponse Module [21] [22]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull; Language model attack\u003c/p\u003e \u003cp\u003e\u0026bull; Adversarial test simple\u003c/p\u003e \u003cp\u003e\u0026bull; Adversarial responding feedback\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull; PII theft and trust loss\u003c/p\u003e \u003cp\u003e\u0026bull; Wrong information\u003c/p\u003e \u003cp\u003e\u0026bull; Reconstruct ML model to perform malicious tasks\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; Attackers make malicious attempts to steal information.\u003c/p\u003e \u003cp\u003e\u0026bull; The input messages are crafted to lead fake response\u003c/p\u003e \u003cp\u003e\u0026bull; An attacker can replicate the response generation module\u0026rsquo;s content to execute malicious attack task without modifying the model parameters\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026bull; Verification Language model\u003c/p\u003e \u003cp\u003e\u0026bull; Hate speech detector (future update)\u003c/p\u003e \u003cp\u003e\u0026bull; \"Network interpretations and data transformation with Style Transformer.\"\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDatabase Module [23]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026bull; SQL injection\u003c/p\u003e \u003cp\u003e\u0026bull; Knowledge graph attack\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026bull; Data deletion, alteration or theft\u003c/p\u003e \u003cp\u003e\u0026bull; Loss of accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; An attacker can submit a harmful SQL statement, which can result in gaining unauthorized access to sensitive data.\u003c/p\u003e \u003cp\u003e\u0026bull; An attacker can manipulate the knowledge graph of a chatbot in order to provide incorrect information.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026bull; Input validation\u003c/p\u003e \u003cp\u003e\u0026bull; Storage mechanism\u003c/p\u003e \u003cp\u003e\u0026bull; Identification detection\u003c/p\u003e \u003cp\u003e\u0026bull; Access control\u003c/p\u003e \u003cp\u003e\u0026bull; Strong encryption\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe comparison of prior work with our developed framework\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChatbot Name\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResearch study topic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePart of Chatbot\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLimitation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOur developed framework\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDeep learning mechanism [24] [25]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChatbots using Deep Learning\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArchitecture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; Take more time for training.\u003c/p\u003e \u003cp\u003e\u0026bull; Costly techniques.\u003c/p\u003e \u003cp\u003e\u0026bull; No Standard Framework.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"5\" rowspan=\"6\"\u003e \u003cp\u003eWe developed a comprehensive blockchain enabled federated learning framework with the integration of fully homomorphic encryption and facial recognition algorithm that can improve privacy / security, accuracy, performance and data management or transparency during three different stages such as login access, data security in transmission, and data privacy/security at the backend of chatbots.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHealth Care chatbot [26] [6]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDesign of Chatbot using Depp Learning\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArchitecture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; Data wasn\u0026rsquo;t trained on time.\u003c/p\u003e \u003cp\u003e\u0026bull; Slow processing\u003c/p\u003e \u003cp\u003e\u0026bull; Lack of security framework\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiagnostic chatbot [27] [28]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDesign and development of diagnostic chatbot for supporting primary healthcare system\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eApplication\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; Less accuracy the human doctor\u003c/p\u003e \u003cp\u003e\u0026bull; Contain small database to store 150 diseases.\u003c/p\u003e \u003cp\u003e\u0026bull; Weak of machine learning part\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePriBot [29] [30]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePriBot: Conversation privacy with Chatbots\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eApplication/Security\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; Risk of malicious users to make the bot useless\u003c/p\u003e \u003cp\u003e\u0026bull; The bot is closer to personal privacy assistant than to a smarter searching interface\u003c/p\u003e \u003cp\u003e\u0026bull; Users less trust on PriBot about their data usage.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChatbots [31] [32]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChatbots: Security, Privacy, Data protection, and Social aspects\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSecurity/Privacy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; Data manipulation on provider side.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpeech review chatbot system [33]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAn reinforcement leaning-based speech censorship chatbot system\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSecurity/Privacy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026bull; The impact of data imbalance and attack detection and aggressive response detection considering multiple rounds of dialogue.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;2 shows that the user's request via message or voice is forwarded through interface to the Node Server to find an appropriate response. Here, the Node Server is in contact with different parts; the environment is the main location for the natural language process for context explanation and has several sub-components; 1 - NLP engine is the main part of the environment that interprets what the user says in conversation time and converts it into structured data for further processing; the intent classifier takes the user request, identifies its meaning, and tracks it back to one of the chatbot supporting intents, the entity extractor extracts the main information of the user request. The Dialog Management Agent manages the current context of the dialog between the user and the chatbot to learn more from the user feedback for user future satisfaction. Question and answer is the main part in the chatbot system that interprets users' frequently asked questions to give users an accurate answer based on its knowledge. This part can give the answer based on two common methods: Manual training, where the company provides a list of questions and answers (Q\u0026amp;A) for the chatbot to find the answer from this list and send it back to the user, and automated training, where the company provides all the related documents and then the chatbot trains through machine learning based on the documents and provides the answer to the user query [34].\u003c/p\u003e"},{"header":"4. CHATBOTS AS A SERVICE","content":"\u003cp\u003eToday\u0026rsquo;s connected world has changed the way of businesses and consumers communication that is commonly known as artificial intelligent applications called chatbots. It is one of the most important forms of communication that can manage the relationship between service providers and customers through a conversational interface such as Facebook, WhatsApp, WeChat, etc., as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e3\u003c/span\u003e; which is integrated using natural language processing [35]. In education, health, industry, banking, and government, chatbots are used to provide real-time communication services to their customers to improve communication and save time and cost because chatbots need to collect and store a large amount of data about the user and the associated conversation [36].\u003c/p\u003e \u003cp\u003eit shows that a user generates a request process through a massaging interface via a speech or text input application and sends it to a natural language parser to convert it into the programming language of the conversation learning engine, or the conversation engine analyses the user request and sends it to the backend of the system, which is connected to many different databases that provide the response to user requests. Most messaging platforms are supported by third-party chatbots and they collect and store a large amount of data related users. Also, they need permission to add more features to enhance the conversation to complete the task on behalf of the user, so the security and privacy of such huge data is a critical issue in the chatbot system, therefore users and businesses need to be aware of data sharing protection [37].\u003c/p\u003e"},{"header":"5.\tCHATBOT SYSTEM SECURITY CHALLENGES","content":"\u003cp\u003eWe have explained the architecture of chatbots in the previous section that chatbots collect and store a large amount of data through communicating with users over the Internet to provide the services, as well as exchange a piece of specific data between user and chatbot system application during the communication. Since chatbots contact with people over the Internet, they are vulnerable to cybersecurity and malicious activities. Therefore, it is very important to investigate data security and confidentiality during transmission and the rest of the system to increase users' confidence in the future use of chatbots [38].\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e5.1 Security and privacy challenges\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eInformation security is the customer's willingness to preserve and control their personal identity information (PII), so privacy protection consists of three stages. First, the collection of personal data, second, control over the data, and third, knowledge of laws and regulations related to data processing and privacy [39]. Chatbots are used in various application areas where a large amount of data needs to be processed. Therefore, security and privacy are critical issues for chatbot systems, especially for the organization that chatbot uses for critical tasks such as financial information, data analytics, etc. Here, we can highlight that secure communication and authentication, data integrity, confidentiality and system availability, transparency and accountability are security challenges for chatbot systems [40]. First of all, it should be ensured that only authenticated users can communicate with chatbots to query/transmit their sensitive information related to financial issue. Second, the integrity and confidentiality of the data must be ensured. The transmitted data can be viewed only by authorized participants and should be protected from any kind of violation and corruption. Third, availability, which protects the system from interruptions and keeps it available to users. Transparency and accountability increase the trustworthiness of the chatbot system. The privacy and security issues arise from the lack of comprehensive security framework to control over the data transferred between the user and the chatbot system [37].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e5.2 AI based chatbots Security threats\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eWith the development of AI technologies, chatbots are able to reproduce human aspects such as conversation through speech, and text with high accuracy like humans. This opens the door to malicious activities and gives a chance to social engineering, man-in-the-middle, and phishing attacks for compromising sensitive information like hackers use bots as tools to imitate humans to trick them into submitting their payment details [40]. The victim unknowingly trusts chatbot requests and does not know that the chatbot is controlled by a cybercriminal group, so hackers collect a large amount of personal data from users through chatbots from numerous conversations with end users every day [41].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e5.3 Chatbots vulnerabilities\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eVulnerabilities are the weaknesses of chatbots that are exploited by adversaries to compromise the security of the system. These vulnerabilities arise from inadequate code, carelessness in code and infrastructure, unprotected and pornographic human errors and make the chatbot system vulnerable to cyber-attacks. Many chatbots use cloud computing services, which have their own threats and vulnerabilities, to store and process the data. So, it is the responsibility of the bot manufacturer to ensure all security processes related to chatbots, who is responsible for restoring the architecture and data flow, which should be encrypted both in transit and at rest in the system environment [42].\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e5.3.1 End-to-End Encryption issue in chatbot\u003c/h2\u003e \u003cp\u003eThis phase performs two functions; first. Transmission of user request to responded agent; second. Transmission user request from the responded agent to the database for providing the information. Communication with chatbot is associated to upper layer of Open System Interconnection (OSI) model so we can concentrate on OSI upper layer (transport-application) [43]. The upper layer of OSI model also face security threats even in some cases the adversaries uses various tools to lunch attacks such as Man-in-the-Middle (MitM) intercepts the legal conversation and modifies it with his malicious message or accessing encrypted data to extract sensitive information [44]. Most of time denial of services attacks also via communication phase that adversary\u0026rsquo;s access to legal conversation and alter it with their own information to participate in communication as legal client for communication disruption with server [45, 46].\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"6.\tDATA COLLECTION AND MANAGEMENT PROBLEMS IN CHATBOTS","content":"\u003cp\u003eChatbots process a huge amount of data and store it in the system, whether the data is structured or unstructured, but data management is a major concern for users because they do not know what the chatbot does with a collected and stored data? How can data be shared without losing integrity and transparency? These are the two most common privacy questions that make users concerned about the security and protection of their data in the system [47]. Security techniques such as encryption, authentication, and verification can improve data protection of such large amounts of data. Therefore, data scientists have recommended and applied natural language processing (NLP) to address and overcome the security challenges in chatbots, because NLP is an artificial intelligence field that can analyse how computers interpret and manage speech and text to collect the information according to the concept of human language to provide a suitable mechanism for computer systems to manage human language and perform various related tasks [48]. Despite the advantages of natural language processing, it is also vulnerable to attack by test-time attackers. These vulnerabilities allow attackers to modify input to address model flaws. The universal attacker tries to find typically ungrammatical phrases to use as inputs for predicting bugs as well as for training-time attacks, such as Poisoning attacks where an attacker injects some malicious codes into the victim\u0026rsquo;s dataset as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003e [49, 50].\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003e explores that adversaries are inject a malicious code or application in dataset while it is training data model and make a backdoor for adversaries to do more malicious activities in the future for effecting users data and send them a triggered data instead of normal and corrected sentences [51].\u003c/p\u003e"},{"header":"7.\tCHATBOTS COMPUTATION PRIVACY ","content":"\u003cp\u003eChatbots are artificial intelligence-based computer programs that can process human conversations both spoken and written and allow people to interact with it as a real person. According to this nature of chatbot, it performs simple tasks, such as answering simple predefined questions and complicated tasks, such as digital assistants. Natural language processing performs all process regardless of chatbots type. On the other hand, computation is the critical part of chatbot and its privacy is very important to process the data and answer the users accurately. Therefore, the developers and owners of chatbots should pay attention to improve the protection of data processing by implementing security and privacy technologies and algorithms [52].We will briefly explain some of them;\u003c/p\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e7.1 Homomorphic Encryption\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eThe collection, handling and processing of sensitive data in chatbots or other AI-powered applications is very important and sensitive for both data and system owners, who must apply strict rules and processing algorithms. Homomorphic encryption is the appropriate and cost-effective solution to ensure confidentiality and prevent unauthorized access to personal and business data [53]. Homomorphic encryption is a collection of encryption algorithms that mimic and implement homomorphic properties, which perform a certain type of operations directly on the encrypted text and provide the same response as on the original message after decryption [54] as depicted in Fig.\u0026nbsp;5. In the field of cryptography, homomorphic encryption is a type of encryption scheme that allows third parties to perform computations on encrypted data. Therefore, encryption is an essential mechanism for maintaining the confidentiality of sensitive information, while conventional encryption mechanisms cannot perform this process on encrypted data [55].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e7.2 Secure Multi-Party computing\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eThis is a technique that allows multiple parties to perform computations together while keeping their input secret. Secure multiparty computation is the solution to various problems in joint computation without compromising data confidentiality, and ensures data confidentiality, independence, and accuracy for all parties involved in the computation \u003cem\u003e[56]\u003c/em\u003e. Secure multiparty computation can perform inference on encrypted data, but is less suitable for training large language models such as chatbots because it includes a single dataset owned by a single entity and SMPC leads to increased computation and communication overhead, which has significant impact on system performance \u003cem\u003e[57]\u003c/em\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e7.3 Federated learning\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eArtificial intelligence applications such as chatbots collect, store, and process massive amounts of data, which impacts performance, computing power and time, quality of service, and most importantly, privacy and security of user data. Due to privacy and security concerns, most data owners and users are not interested in sharing data with chatbots because chatbots are data-driven applications and collect user data from text, voice, or multimedia conversations [58].Privacy and security of user data during training of deep learning models may be violated by shared datasets. As mentioned earlier, there are many approaches to data security and privacy, but all of them require access to user data. In contrast, federated learning is an approach to improve data security and privacy where data is not collected in a central location, but the user\u0026rsquo;s data remains in its own location. Federated learning update the information through transferring deep leaning model which is trained over client\u0026rsquo;s private dataset as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e6\u003c/span\u003e, this way federated learning protects data privacy and security [59, 60].\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e7.4 Blockchain\u003c/h2\u003e \u003cp\u003eBlockchain technology employs a distributed database and ledger to securely timestamp and record transactions in a system. The primary idea behind blockchain technology is to provide a secure environment for anyone who communicate for exchanging the information through a public connection [61]. Blockchain operates on a peer-to-peer network where each node having a permanent and immutable copy of the ledger that keeps track of the entire map and the number of chained blocks using a hash technique [62, 63]. It is commonly used to securely transport or exchange information over a network. Unlike traditional centralized systems, which are controlled by a single entity, blockchain is a decentralized network of computers (nodes) where, each node keeps a copy of all transactions. If a new user or node wants to join and create the block of transactions to the system, then various consensus procedures like proof of work and proof of stake are used to validate and provide the agreement of network nodes (miners) about new adding node and block for transactions [64, 65]. Blockchain employs cryptographic techniques to safeguard transactions and regulate the generation of new units as depicted in Fig.\u0026nbsp;7. Public and private keys are used to secure transactions and restrict access to the blockchain. Once a block is uploaded to the blockchain, it is exceedingly impossible to change or delete the data contained within it. The blockchain\u0026rsquo;s immutability is achieved through cryptographic hashing and the consensus method, resulting in a tamper-resistant ledger [49]\u003c/p\u003e"},{"header":"8.\tADDRESSING THE SECURITY AND PRIVACY CHALLENGES IN CHATBOTS SYSTEMS","content":"\u003cp\u003eAs mentioned earlier that chatbots are automated computer application which effected people live in various aspects like providing personal assistance, information, offering services and so on. Whenever a human-like dialog system has been developed known as chatbots the attention must be paid to potential security and privacy challenges, vulnerabilities that can lead data leaked and exploited. There are various type of security and privacy challenges that affect different parts of chatbot according to its architecture [67]. To better understand we have divided chatbots architecture into three parts where every part experience various security and privacy challenges with their own solutions;\u003c/p\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e8.1 Access control and Authentication\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eThe widespread uses of chatbots increases security and privacy risk especially for financial service providers that is harmful for both users and system. First layer to improve security/privacy of chatbot is to control login access to prevent illegal login to chatbot [31]. Access control can improve data integrity and confidentiality because it strongly restricts unauthorized users access to chatbots [68]. Therefore to have strong login access control we recommend a layer combined from username/password and facial recognition because it allow the chatbot to be connected with only with authorized and legal user who will also be accountable for his/her future actions [69].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e8.2 Data Transmission Security/Privacy\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eArtificial intelligence based chatbots communicate with users and exchange data over a public internet connection. End-to-end encryption is a type of communication where only the legal and authorized parties (source and destination) can see, read and decrypt the message without anyone else. Therefore it is very important to secure the communication between user and chatbot to ensure that adversaries cannot access user data during transmission but only legal user can user encrypt and decrypt keys to read the messages [70]. We would like to use a new and secure algorithm (full Homomorphic Encryption) to improve the security of data during transmission because it allow the process to be performed over encrypted data in the contrast of other encryption methods and algorithms.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTraditional encryption algorithms spread the concept of distributed keys where public and private keys are exchanging over internet during communication because they need process over decrypted data. In contrast, full homomorphic encryption algorithm (FHE) allows a complex mathematical operation to be performed on encrypted data during transmission without decryption [71]. FHE consist of three steps for data encryption process; key generation, encryption and decryption. Key generation step is randomized that take security parameters as input and generate public (PK) and secret key (SK) to encrypt data. Encryption step also randomized that takes PK and plaintext as input and generates the cihpertext, while decryption needs SK and ciphertext as input and generated plaintext [72], as depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e6\u003c/span\u003e. Even though the adversaries access to data during transmission but they cannot decrypt and read the information therefore it reduce the risk of man-in-the-middle attack. Form another hand, full homomorphic encryption can support encrypted image processing, For example chatbot get the image from user and forward it to deep learning model for further processing and the result is transmitted back to chatbot to be displayed for user accordingly [73]. So the implementation of FHE can improve data security/privacy during transmission between user and chatbot [74].\u003c/p\u003e\u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e8.3 Data Storage and Security in Backend\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eAs earlier mentioned that chatbot collect a huge amount of data to a single point, so the processing of such huge data is challengeable and big issue. Chatbot are used in various application domain so it needs to keep data integrity, confidentiality, transparency, and accountability. From another hand, users also concerned about their data that how chatbot deal with it and how it share the data by keeping integrity and transparency [75]. To address these challenges and users concerns we developed a framework that integrated blockchain enabled federated learning (see section \u003cspan refid=\"Sec22\" class=\"InternalRef\"\u003e8.4\u003c/span\u003e) to improve data security/privacy and transparency in the backend of chatbot. Blockchain is a distributed ledger technology that maintains a continuously growing ledger on the network to provide a secure transaction with a timestamp recording mechanism to support data security and privacy [76, 77]. Ledger is systematic system for data structure that consist of many blocks chained by a cryptographic mechanism. Therefore it uses a chained blocks to store and transmit data with public and private keys to verify and sign the transactions because it calls blockchain [78]. Ledger technology make the smart contract and its execution immutable, irreversible, and undesirable. In addition blockchain provide data persistence, distributed data control, data management, transparency and accountability. Blockchain is used and deployed in various fields to provide a secure environment for data processing. Recently, it can also be used with natural language processing because blockchain stores and share the data with other users through a distributed ledger [79]. Blockchain is used to store data as a smart contract in natural language processing to improve storage mechanism for trustworthiness and prevent SQL injection attack which are important for secure computation. In addition blockchain can store sentiment analysis data as smart contract in natural language processing to allow the users to access it as open source rather than closed or preparatory source that is cheap and easy for small businesses [80]. Blockchain helps the artificial intelligent system to process the data accurately and effectively by using smart contracts to connect different databases. Therefore, it makes data analysis true and the decision-making process becomes better for the organization. Therefore, Blockchain NLP allows users to compose a text through their voice or simply type the text using the keyboard, which is processed by artificial intelligent model to protect it from language model attacks and malicious injections [81, 82].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e\u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e8.4 Proposed Blockchain enabled fusion chatbots Security/Privacy framework\u003c/span\u003e\u003c/h2\u003e \u003cp\u003eThis framework has been developed to improve chatbots security for facilitating some security functions to perform critical functionalities in a secure manner. The framework integrates blockchain-based federated learning and homomorphic encryption with face recognition that enhance the security/privacy of three correlative phases; (1)- Secure login access (2)- secure data transmission (3)- Data security, privacy and transparency in chatbots backend. Obviously the framework included two main functions front end for login access that user have direct access and backend for data storage and transactions but the third part is operating between frontend and backend for secure data transmission. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e9\u003c/span\u003e shows the main framework that has two main parts front end and backend where the user has direct access to front end for performing the required options and backend stores transactions data. Figure\u0026nbsp;10 shows the flow diagram that describes the order and relationship between frond and backend parts and Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e11\u003c/span\u003e shows the sequence diagram of main framework to explain the operation and interaction between users and chatbot. For framework implementation needs various datasets for chatbots such as NewQA dataset for questions and answer that contains over 10000 human generated questions and answers. This data set use Reading Comprehensive Model (RCM) to understand and interpret new articles and investigate new methods for handling complex questions and large documents. The Scheme Guided Dialog (SGD) dataset is used for dialog interactions. It consist of over 20k annotated multi-domain, task-oriented conversations between a human and virtual assistant. We would like to explain the structure of every part separately;\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003e8.4.1 Research and evaluation method\u003c/h2\u003e \u003cp\u003eThis research paper includes both theoretical and experimental method to evaluate some parts like the performance, accuracy and privacy of the proposed framework. First we performed a systematic research through various digital libraries study and carry out related research topics to shape a conceptual framework. Second, the developed framework tested in a chatbot virtual environment called RASA which is open source framework based on machine learning to create highly accurate text and voice based conversation chatbots. RASA action server provides the environment to write a code in python version 3.6 that mainly used to trigger actions of some parts like federated learning and FHE based on hyper ledger fabric with a related data sets mentioned in related parts.\u003c/p\u003e\u003cp\u003eFirst. Secure login access: the user who wants to connect to chatbot he/she needs to be authenticated through first layer of security check which is included username/password and face recognition. Because it allows the chatbot to be connected only to a specific and authorized user who have access to sensitive data. He/she will also be accountable for their actions in the future, therefore it can improve data integrity and confidentiality by restricting unauthorized user access. The framwork used Viola-Jones algorithm for face detection with all features to train the classifier and we recommended convolutional neural network (CNN) which is a type of artificial neural network for picture classification, segmentation, processing and accuracy. CNN can extract the features from higher layers data to recognize image and it has the ability to develop internal representation of two dimensions of the image that allow the model to learn the positon and scale of the image. For the implementation it uses the Tufts face dataset which is a large-scale and public benchmark for face recognition. It is included 10000 image for different countries with age range form 4\u0026ndash;74, the process are explained in Figs.\u0026nbsp;12, 13, and 14.\u003c/p\u003e\u003cp\u003eSecond. Secure Data Transmission: chatbots and user communicate through a public internet connection so it needs to be secure from unauthorized access to understand or alter the data, only the involved parties should be able to encrypt and decrypt the messages. Therefore we recommended a full homomorphic encryption algorithm for secure transmission because as we mentioned earlier. FHE allows the computation process to be performed on encrypted data without decryption. The computation process are explained in algorithm1.\u003c/p\u003e \u003cp\u003eWe tested it in a virtual environment including 5 computational clients were used to calculate the encryption time according to file size, which is performed on plaintext from 256\u0026ndash;1792 bits data by adding 256 bits in each round. The Fig.\u0026nbsp;15 shows that fully homomorphic algorithm encrypt more data in less amount of time but its encryption time increased along with the size of plaintext.\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e16\u003c/span\u003e shows the fully homomorphic algorithm performance time for encryption and decryption according to keys size. Here the key sizes are selected as 512, 1024, 1536, 2048, 2560 bit to achieve 80,112,128,192,256 bit security levels. The encryption and decryption time has direct relationship to key size because of exponentiation operation that is increasing exponentially.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eThird.\u003c/b\u003e Data Security/privacy at backend: this is the big and important part of chatbots for data mining and processing where the user has access indirectly to ensure security and privacy as its flow diagram depicted in Fig.\u0026nbsp;10. We integrated blockchain enabled federated learning in framework for improving data security, privacy, transparency, accuracy and performance of chatbot. Federated learning is adequate mechanism to address such problem because federated learning use or exchange local and global model to update the information instead of collecting a huge amount of real data in a single central location. They only need to train local model over their personal dataset and send it to central server for aggregation to create global model. If we consider there are k clients over which the data has partitioned with \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({P}_{k}\\)\u003c/span\u003e\u003c/span\u003e as data set of indexes on \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\\)\u003c/span\u003e\u003c/span\u003e clients, so on each client \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({n}_{k}=\\left|{p}_{k}\\right|.\\)\u003c/span\u003e\u003c/span\u003eAnd local model training is conducting using local dataset as written in Eq.\u0026nbsp;1\u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equa\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$${w}_{t}^{k}={w}_{t-1}^{k}-\\nabla wf\\left({w}_{t-1}^{\\left(c\\right)}\\right)$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003eAfter local update it will be shared with server for aggregation, the process is shown in equations 2.\u003c/p\u003e\u003cp\u003e \u003cdiv id=\"Equb\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$${F}_{k}\\left(w\\right)=\\frac{1}{{}_{k}}\\sum _{i={\\rho }_{k}}fi\\left(w\\right)$$\u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Equc\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$gk=\\nabla {F}_{k}\\left({w}_{t}\\right)$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003cbr\u003e\u003c/p\u003e \u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({w}_{t}\\)\u003c/span\u003e\u003c/span\u003e shows model weights in communication round \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(t\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({w}_{t}^{k}\\)\u003c/span\u003e\u003c/span\u003e indicates model weights on communication round \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(t\\)\u003c/span\u003e\u003c/span\u003e on client \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\\)\u003c/span\u003e\u003c/span\u003e, learning rate has shown by η, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({p}_{k}\\)\u003c/span\u003e\u003c/span\u003e shows set of data points on client \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\\)\u003c/span\u003e\u003c/span\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({}_{k}\\)\u003c/span\u003e\u003c/span\u003e shows number of data points on client \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(k\\)\u003c/span\u003e\u003c/span\u003e. The Eq.\u0026nbsp;3 shows that central server aggregates all these gradients to apply the updates.\u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equd\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$${w}_{t+1}\\leftarrow {w}_{t}- \\sum _{k=1}^{k}\\frac{{n}_{k}}{n} gk$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003eCollecting a huge amount of data to a single point will be risk full from the prospective of privacy and security and it can also affect performance and accuracy of computation. Figure\u0026nbsp;17 show the accuracy of federated learning (distributed) versus centralized system that FL has improved the accuracy of computation till 90 percent but the accuracy in centralized system is decreasing when the iterations get high (75 percent). Figure\u0026nbsp;18 shows the difference of performance between centralized and federated learning system that the federated learning (distributed) has high performance than centralized.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eDespite of this advantage of federated learning, there are still some concerns about data lose because federated learning operates centrally therefore the attackers can access to data trained models or can attack on central server for data leakage and access. Therefore we integrated blockchain technology to address the challenges of federated learning. Blockchain is more secure technology because it operates in decentralization manner. It stores all data on block-based manner, therefore once smart contract creates the block according to the consensus algorithm then it will not be changeable so it is more affective to improve data CIA triad. The system employs two smart contracts to manage user requests. The first contract, provided by the system, handles user registration and login. The second contract, on the other hand, is provided by the organization and is responsible for the encoding of the business logic for data transactions. For instance, in the case of a bank system, the smart contract simulates the functionalities of a financial institution. As such, every user of the system is expected to hold an account with the bank and perform financial activities such as balance inquiries and money transfers, which are carried out by the bank within the system. After conducting a thorough analysis of various public and private blockchains for deployment, we have concluded that public blockchains are more secure. However, they have some drawbacks such as being slow, open to all, and incurring significant costs to process and store data in a smart-contract like Ethereum. Therefore, we have decided to use a private blockchain system. Currently, Hyperledger Fabric is the most stable and popular private blockchain platform that supports smart contract functionality. This platform offers a unique concept of a channel that enables multiple blockchains to be managed within a single network. This creates a layer of confidentiality between various organizations to ensure that different activities remain private between different entities. Hyperledger Fabric uses several network entities like node, creator and validator showed in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e9\u003c/span\u003e. A smart contract in Fabric is referred to as chain and can be invoked through transactions. Node submitting a transactions to validator which is responsible to check the validation of an entity for its permission to perform an activity in a ledger encoded during transaction. The validation transactions status forwarded to a creator to create a new block according to transaction and return it to validation and node entity to update the ledger in blockchian as Fig.\u0026nbsp;19 Shows that transaction process has improved with high speed in our scheme from the prospective of user\u0026rsquo;s creation and transaction.\u003c/p\u003e \u003cp\u003eThe developed framework also includes sentiment analysis beside of technical and security parts that can evaluate customer interactions with chatbots. We used a lexicon based sentiment analysis techniques to extract the emotional polarity from the chat or text or users have with chatbots [83]. It examines all the words and sounds used by the customer during the interaction with the chatbot to understand the user's state, whether he is happy, sad, angry, or we can divide all these status on three major categories such as positive, negative and neutral status [84], Fig.\u0026nbsp;20 shows the flow diagram and algorithm2 the steps of sentiment analysis algorithm. Therefore, sentiment analysis can identify the areas that need to be improved. On the other hand, sentiment analysis is a form of cybersecurity perspective that refers to natural language processing and AI-based techniques to analyse users' state, attitudes, and opinions expressed in text and speech related to cybersecurity issues. Gathering information in the form of sentiment analysis that provides insight into how users think about services, specific products, or security issues is therefore key to improving the quality of services and security [85].\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"1\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAlgorithm 2: Sentiment Analysis Algorithm\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eINPUT\u003c/b\u003e: Text chat ᴛ, the sentiment lexicon Ⅼ.\u003c/p\u003e \u003cp\u003e\u003cb\u003eOUTPUT\u003c/b\u003e: S\u003csub\u003emt\u003c/sub\u003e = {P, Ng, or N} and Straight S, where P: Positive\u003c/p\u003e \u003cp\u003eNg: Negative, N:Neutral\u003c/p\u003e \u003cp\u003e\u003cb\u003eINITIALIZATION\u003c/b\u003e: SumPos and SumNeg\u0026thinsp;=\u0026thinsp;0, where\u003c/p\u003e \u003cp\u003eSumPos: accumulates the polarity of positive tokens \u003cem\u003et\u003c/em\u003e\u003csub\u003e\u003cem\u003ei\u0026minus;smt\u003c/em\u003e\u003c/sub\u003e in T\u003c/p\u003e \u003cp\u003eSumNeg: accumulates the polarity of negative tokens \u003cem\u003et\u003c/em\u003e\u003csub\u003e\u003cem\u003ei\u0026minus;smt\u003c/em\u003e\u003c/sub\u003e in T\u003c/p\u003e \u003cp\u003e\u003cb\u003eBegin\u003c/b\u003e\u003c/p\u003e \u003cp\u003e1. For each t\u003csub\u003ei\u003c/sub\u003e \u0026isin;\u003cem\u003eT\u003c/em\u003e do\u003c/p\u003e \u003cp\u003e2. Search for t\u003csub\u003ei\u003c/sub\u003e in \u003cem\u003eL\u003c/em\u003e\u003c/p\u003e \u003cp\u003e3. If ti \u0026isin; Pos-list \u003cem\u003ethen\u003c/em\u003e\u003c/p\u003e \u003cp\u003e4. SumPos SumPos\u0026thinsp;+\u0026thinsp;t\u003csub\u003ei-smt\u003c/sub\u003e\u003c/p\u003e \u003cp\u003e5. Else if ti \u0026isin; Neg-list \u003cem\u003ethen\u003c/em\u003e\u003c/p\u003e \u003cp\u003e6. SumNeg SumNeg\u0026thinsp;+\u0026thinsp;t\u003csub\u003ei-smt\u003c/sub\u003e\u003c/p\u003e \u003cp\u003e7. End if\u003c/p\u003e \u003cp\u003e8. End for\u003c/p\u003e \u003cp\u003e9. If SumPos \u0026gt; |SumNeg| \u003cem\u003ethen\u003c/em\u003e\u003c/p\u003e \u003cp\u003e10. S\u003csub\u003emt\u003c/sub\u003e = P\u003c/p\u003e \u003cp\u003e11. S\u0026thinsp;=\u0026thinsp;SumPos/ (SumPos\u0026thinsp;+\u0026thinsp;SumNeg)\u003c/p\u003e \u003cp\u003e12. Else if SumPos \u0026lt; |SumNeg| \u003cem\u003ethen\u003c/em\u003e\u003c/p\u003e \u003cp\u003e13. Smt\u0026thinsp;=\u0026thinsp;Neg\u003c/p\u003e \u003cp\u003e14. S\u0026thinsp;=\u0026thinsp;SumNeg / (SumPos\u0026thinsp;+\u0026thinsp;SumNeg)\u003c/p\u003e \u003cp\u003e15. Else\u003c/p\u003e \u003cp\u003e16. Smt\u0026thinsp;=\u0026thinsp;N\u003c/p\u003e \u003cp\u003e17. S\u0026thinsp;=\u0026thinsp;SumPos/ (SumPos\u0026thinsp;+\u0026thinsp;SumNeg)\u003c/p\u003e \u003cp\u003e18. End If\u003c/p\u003e \u003cp\u003e\u003cb\u003eEnd\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"9. CHATBOTS VS CHATGPT ","content":"\u003cp\u003eChatbots are an integral part of the digital world and are revolutionizing the way businesses interact with their customers. Therefore, the technological capabilities of chatbots are improving daily, from rule-based to complex conversational agents driven by artificial intelligence and machine learning algorithms. Although chatbots and ChatGPT are both conversational artificial intelligence technologies that have become increasingly popular in recent years, rises some security challenges and both are conversational agents that communicate with humans through natural language processing techniques [86, 87]. However, there are some minor differences between them in terms of structure, obtaining information, and generating responses to user queries as depicted in Table \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Chatbots are computer program which is designed to mimic human conversation through text or voice and are an application of artificial intelligence that creates an environment for communication between humans and machines in a conversational manner [32]. Chatbots focus on a specific domain, so they learn from decision trees or data predefined by the owner, as well as from user interactions. AI chatbots use natural language understanding and processing to generate human-like conversations. ChatGPT is a generative, pre-trained language model for chatbots developed by openAI. ChatGPT can handle a wide range of topics and domains because it uses deep leaning and a transform architecture to train the model through a vast amount of textual data on the web, enabling it to understand and generate human-like conversations [88].\u003c/p\u003e"},{"header":"10. CONCLUSION ","content":"\u003cp\u003eIn this paper, we conducted a survey research about chatbots security and privacy challenges. Some security threats, vulnerabilities and challenges existed to compromise chatbot privacy. Chatbot security and privacy is compromising via three stages: first. Login access, data during transmission and data at the backend, because chatbot always in contact with users to exchange and it collects a huge amount of data to be processed. As we analysed and investigated chatbot security system, the owners of chatbots only concentrated on communication and service providing aspect rather than security and privacy. Although some of the chatbots have security measures but they used very low level techniques that hackers can exploit it easily to access user\u0026rsquo;s data. Hence we developed a comprehensive and fusion framework to improve the privacy, security, accuracy, performance and encryption process of chatbot in all above mentioned security stages\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eChatbot comparison with ChatGPT\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAttributes\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAI-Chatbots\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eChatGPT\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArchitecture and design\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMachine learning model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGenerative Pre-trained Transformer\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFlexibility\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFlexibility based on predefined rule\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh flexibility\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrained on specialized dataset\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePre-trained on vast internet based data\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConversational Depth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOffer depth based on training data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOffer more depth\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePersonalization\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCan make personalized suggestions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePersonalization is extended\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLearning capabilities\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLearning from specific training data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLearn from vast amount training data\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUse cases\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTask automation, customer support and information retrieval\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCreative writing tools, virtual assistance, chat experience\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSecurity and privacy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVulnerable to data breaches\u003c/p\u003e \u003cp\u003e\u0026minus; User data collection\u003c/p\u003e \u003cp\u003e\u0026minus; Data leak\u003c/p\u003e \u003cp\u003e\u0026minus; Sharing confidential data\u003c/p\u003e \u003cp\u003e\u0026minus; Algorithmic bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVulnerable to data breaches\u003c/p\u003e \u003cp\u003e\u0026minus; spread malicious software\u003c/p\u003e \u003cp\u003e\u0026minus; business email compromise\u003c/p\u003e \u003cp\u003e\u0026minus; user data collection\u003c/p\u003e \u003cp\u003e\u0026minus; sharing confidential data\u003c/p\u003e \u003cp\u003e\u0026minus; phishing emails\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eduring communication. This framework is integrated three fusion system: first. Face recognition system beside of normal authentication method (username and password) for secure and authorized login to prevent attacks related to login access. Second; A full homomorphic encryption algorithm for encrypted and secure transmission between user and chatbot to prevent malicious attacks like DoS and MiTM and access during transmission. Third. Blockchain enabled federated learning for data privacy, security and transparency at the backend of chatbots. The experimental test showed, the implementation of this framework improved accuracy and performance as well as prevent unauthorized users and attackers to exploit chatbot security while exchanging data during all above mentioned stages and it ensure the system that information cannot be shared with others. All conversations are transferred via a secure network and transactions are signed digitally by user private key to ensure the confidentiality of data and transaction without validated private key signature will not be processed, In addition, blockchain enabled FL is a distributed platform and protect the\u003c/p\u003e \u003cp\u003esystem against DoS and MiTM attacks. So we can summarize achievement of contributions in Table \u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e:\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cem\u003esummary table of contributions (\u003c/em\u003esection \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e1.1\u003c/span\u003e\u003cem\u003e)\u003c/em\u003e \u003cspan type=\"BoldSmallCaps\" class=\"BoldSmallCaps\" name=\"Emphasis\"\u003eReferences\u003c/span\u003e\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eObjectives\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSolutions\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSecure Login Access\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTo improve login access in a chatbot we developed a framework that is included username/password and facial recognition because first time a user should register his/her face with chatbots and store the relevant information in a blockchain secure technology that is hard to be altered or tracked by unauthorized users. In addition we used CNN algorithm to process and classify all features related to picture for accuracy. Therefore facial recognition enable the chatbot to be connected with authorized user and restrict all illegal login access.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSecure Data Transmission\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTo have secure data transmission we have used fully homomorphic encryption algorithm because it allows the process to be performed faster over encrypted data in the contrast of other encryption algorithms. Only the source and destination are known about encryption and decryption key for data reading. Therefore adversaries can\u0026rsquo;t access data during transmission.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eData Security/Privacy and Transparency in Backend of Chatbot\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTo improve data privacy and trust on using chatbot we used federated learning where users only need to exchange trained model with central location instead of transferring the primary data for further process and chatbots functionalities. In addition we integrated blokchain with federated learning system in backend of chatbot to be responsive to the concerns of data transparency, security, confidentiality, performance and sharing with other users or parties. Because blockchain store the data in a block based and distributed manner which is include hash functions of previous block, therefore once the data transacted and stored in blockchain then no one can change because all participated nodes has the transaction copy.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"11. FUTURE DIRECTION ","content":"\u003cp\u003eFor future direction we recommend to enable chatbot especially financial chatbot based on beyond 5 generation mobile technology which is envisioned to revolutionize chatbots for delivering different services in the future. In addition I recommend to extend various chatbots such as social-bots testing against other vulnerabilities.\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e all data sets analysed during this study to support the results of the article are publicly available.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions:\u0026nbsp;\u003c/strong\u003eauthors contributed equally to the conceptualization and design of the solution for mentioned challenges. Data collection and analysis performed by Nasir Ahmad Jalali and Professor Chen Hongsong provide supervision as well as reviewed the paper for quality improvement. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAcknowledgment\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis work was supported by the National Key Research and Development Program of China (No.2023YFC3303803 and 2023YFC3303800), State Key Laboratory of Public Big Data of Guizhou University (No.PBD2023-24), CCF NSFocus Kunpeng Foundation (No.CCF-NSFocus2023012) ，Fundamental Research Funds for the Central Universities (No. FRF-AT-19-009Z and FRF-AT-20-11) from the Ministry of Education of China.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no financial or non-financial interests to disclose or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eC. B. N. \u0026amp;. C. A. B. Asbjorn Folstad, \u0026ldquo;What Makes Users Trust a Chatbot for Customer Services? An Exploratory Interview Study,\u0026rdquo; in \u003cem\u003eThe proceeding fof the Fifth International conferenceon Internet Science\u003c/em\u003e, Oslo, Norway, 2018.\u003c/li\u003e\n \u003cli\u003eS. D. S. M. A. B. Lorenz Cuno Klopfenstein, \u0026ldquo;The Rise of Bots: A Survey of Conversational Interfaces, patterns, and Paradiams,\u0026rdquo; in \u003cem\u003eDIS17: Proceedings of the 2017 Conference on Designing Interactiv System\u003c/em\u003e, 2017.\u003c/li\u003e\n \u003cli\u003eP. Gentsch, \u0026ldquo;Conversational AI: How (Chat)Bots Will Reshape the Digital Experience,\u0026rdquo; in \u003cem\u003eAI in Marketing Sales and Services: How Marketers without a Data Science Degree can user AI, Big Data and Bots\u003c/em\u003e, Frankfurt, Germany, SpringerLink, 2019, pp. 81-95.\u003c/li\u003e\n \u003cli\u003eM.-T. L. D. R. S. e. a. Daniel Adiwardana, \u0026ldquo;Towards a Human-like Open-Domain Chatbot,\u0026rdquo; \u003cem\u003earXiv:2001.09977v3 [cs.CL] 27 Feb 2020,\u0026nbsp;\u003c/em\u003e2020.\u003c/li\u003e\n \u003cli\u003ea. S. S. Takuma Okuda, \u0026ldquo;AI-Based Chatbot Service for Financial Industry,\u0026rdquo; \u003cem\u003eFUJITSU Scientific and Technical Journal ,\u0026nbsp;\u003c/em\u003evol. 54, no. 2, pp. 4-8, 2018.\u003c/li\u003e\n \u003cli\u003em. P. Milind H Shah, \u0026ldquo;Theoretical Evaluation of Securing Modules for Educational Chatbot,\u0026rdquo; in \u003cem\u003eProceedings of the Sixth International Conference on Intelligent Computing and Control System (ICICCS)\u003c/em\u003e, 2022.\u003c/li\u003e\n \u003cli\u003eN. A. S. A. a. K. A. Mohammad Amin Kuhail, \u0026ldquo;Interacting with educational chatbots: A systematic review,\u0026rdquo; \u003cem\u003eEducation and Information Technologies,\u0026nbsp;\u003c/em\u003evol. 28, pp. 973-1018, 2023.\u003c/li\u003e\n \u003cli\u003eh. A. J. P. a. V. S. Asim Mohammed Eltahir, \u0026ldquo;Review of Chatbot Security Systems,\u0026rdquo; in \u003cem\u003e2022 26th International Conference on Circuits, Systems, Communications and Computers (CSCC)\u003c/em\u003e, 2022.\u003c/li\u003e\n \u003cli\u003eJ. R. N. a. D. H. Rahime Belen Saglam, \u0026ldquo;Privacy Concerns in Chatbot Interactions: When to Trust and When to Worry,\u0026rdquo; in \u003cem\u003e23rd HCI International Conference, HCII 2021\u003c/em\u003e, Swithzerland, 2021.\u003c/li\u003e\n \u003cli\u003eM. I. S. S. H. e. a. Nazar Waheed, \u0026ldquo;An Empirical Assessment of Security and Privacy Risks of Web-Based Chatbots,\u0026rdquo; \u003cem\u003eWISE international Journal,\u0026nbsp;\u003c/em\u003epp. 1-25, 2022.\u003c/li\u003e\n \u003cli\u003ea. K. B. Erik Derner, \u0026ldquo;Beyond the Safguards: Exploring the Security Risks of ChatGPT,\u0026rdquo; \u003cem\u003earXiv:2305.08005v1 ,\u0026nbsp;\u003c/em\u003e2023.\u003c/li\u003e\n \u003cli\u003eM. B. Hoy, \u0026ldquo;Alexa, Siri, Cartana, and More: An Introduction to Voice Assistants,\u0026rdquo; \u003cem\u003eMedical Reference Services Quarterly ,\u0026nbsp;\u003c/em\u003evol. 37, no. 1, pp. 81-88, 2018.\u003c/li\u003e\n \u003cli\u003eF. C. M. L. e. a. Fabio Clarizia, \u0026ldquo;Chatbot: An Education Support System For Students,\u0026rdquo; in \u003cem\u003eInternational Symposium on Cyberspace Safety and Security\u003c/em\u003e, 2018.\u003c/li\u003e\n \u003cli\u003e\u0026amp;. A. A.-I. Chinedu Wilfred Okonkwo, \u0026ldquo;Chatbots Applications in Education: A Systematic Review,\u0026rdquo; \u003cem\u003eComptuers and Education: Artificial Intelligence,\u0026nbsp;\u003c/em\u003evol. 2, 2021.\u003c/li\u003e\n \u003cli\u003eX. Huang, \u0026ldquo;Chatbot: Design, Architecture, and Applications,\u0026rdquo; Univesity of Pennsylvania: School of Engineering and Applied Science , Pennsylvania, 2021.\u003c/li\u003e\n \u003cli\u003eDasagrandhi, \u0026ldquo;Understanding The Conversational Chatbot Architecture,\u0026rdquo; V-Soft Consulting, 2550 Eastpoint Pkwy Suite 300 Luisville, Ky 40223, 2020.\u003c/li\u003e\n \u003cli\u003eD. J. M. L. e. a. Jing Xu, \u0026ldquo;Bot-Adversarial Dialogue for Safe Conversational Agents,\u0026rdquo; in \u003cem\u003eProceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies\u003c/em\u003e, New York, 2021.\u003c/li\u003e\n \u003cli\u003eN. M. e. a. Domna Bilika, \u0026ldquo;Hello Me, Meet the Real Me: Audio Deepfake Attacks on Voice Assistants,\u0026rdquo; \u003cem\u003earXiv:2302.10328v1 [cs.CR],\u0026nbsp;\u003c/em\u003e2023.\u003c/li\u003e\n \u003cli\u003eM. L. J. V. a. S. L. Hyunji Chung, \u0026ldquo;Alexa, Can I Trust You?,\u0026rdquo; \u003cem\u003eIEEE Xplore,\u0026nbsp;\u003c/em\u003evol. 9, pp. 100-105, 2017.\u003c/li\u003e\n \u003cli\u003eS. B. a. P. Z. Krishna Gondaliya, \u0026ldquo;SLA as Mechanism to Manage Risks Related to Chatbot Services,\u0026rdquo; in \u003cem\u003e2020 IEEE 6th Intl Conference on Big Data Security on Cloud (BigDataSecurity)\u003c/em\u003e, 2020.\u003c/li\u003e\n \u003cli\u003eH. Q. G. W. e. a. Cheng Qian, \u0026ldquo;Distinguish Sens from Nonsens:Out-of-Scope Detection for Virtual Assistants,\u0026rdquo; \u003cem\u003earXiv:2301.06544v1 [cs.CL] 16 Jan 2023,\u0026nbsp;\u003c/em\u003e2023.\u003c/li\u003e\n \u003cli\u003eX. F. Z. X. e. a. Yang Zheng, \u0026ldquo;Why Adversarial Reprogramming works, when it fails, and how to tell the difference,\u0026rdquo; \u003cem\u003eInformation Sciences,\u0026nbsp;\u003c/em\u003evol. 632, pp. 130-143, 2023.\u003c/li\u003e\n \u003cli\u003eF. S. a. C. T. Stefano Bistarelli, \u0026ldquo;A Chatbot Extended with Argumentation,\u0026rdquo; in \u003cem\u003e5th workshop on Advances In Argumentation In Artificial Intelligence\u0026nbsp;\u003c/em\u003e, Milano, Italy, 2020.\u003c/li\u003e\n \u003cli\u003eA. D. V. K. Neeraj Singh Kadayan, \u0026ldquo;Chatbot Using Deep Learning,\u0026rdquo; \u003cem\u003eJournal of Emerging Technologies and Innovative Research (JETIR),\u0026nbsp;\u003c/em\u003evol. 6, no. 2, pp. 137-141, 2019.\u003c/li\u003e\n \u003cli\u003eS. Z. E. e. a. Hana Emma Wube, \u0026ldquo;Text-Based Chatbot in Financial Sector: A Systematic Literature Review,\u0026rdquo; \u003cem\u003eData Science in Finance and Economics ,\u0026nbsp;\u003c/em\u003evol. 2, no. 3, pp. 209-236, 2022.\u003c/li\u003e\n \u003cli\u003eS. S. P. N. G. P. V. Nivila A, \u0026ldquo;Design of Chatbots Using Deep learning,\u0026rdquo; \u003cem\u003eInternational Research Journal of Engineering and Technology (IRJET) ,\u0026nbsp;\u003c/em\u003evol. 9, no. 20, pp. 1105-1110, 2022.\u003c/li\u003e\n \u003cli\u003eN. R. Bushra Kidwai, \u0026ldquo;Design and Development of Diagnostic Chatbot for supporting Primary Health Care System,\u0026rdquo; \u003cem\u003eProcedia Computer Science ,\u0026nbsp;\u003c/em\u003evol. 167, pp. 75-84, 2020.\u003c/li\u003e\n \u003cli\u003ea. S. D. Aishwarya Surani, \u0026ldquo;Understanding Privacy and Security Postures of Healthcare Chatbots,\u0026rdquo; \u003cem\u003eAssociation for Computing Machinery (ACM),\u0026nbsp;\u003c/em\u003e2022.\u003c/li\u003e\n \u003cli\u003eK. F. K. G. S. K. A. Hamza Harkous, \u0026ldquo;PriBots: Conversational Privacy with Chatbots,\u0026rdquo; in \u003cem\u003eWorkshop on the Future of Privacy Indicators, at the Twelfth Symposium on Usable Privacy and Security (SOUPS)\u003c/em\u003e, Denver, Colorado , 2016.\u003c/li\u003e\n \u003cli\u003eA. R. M. S. F. e. a. Md. Saiful Islam Bhuiyan, \u0026ldquo;BONIK: A Blockchain Empowered Chatbot for Financial Transactions,\u0026rdquo; in \u003cem\u003eConference on Trust, Security and Privacy in Computing and Communications\u0026nbsp;\u003c/em\u003e, 2020.\u003c/li\u003e\n \u003cli\u003eJ. N. e. a. Martin Hasal, \u0026ldquo;Chatbots: Security, Privacy, Data protection, and Social aspects,\u0026rdquo; \u003cem\u003eWiley,\u0026nbsp;\u003c/em\u003epp. 1-13, 2021.\u003c/li\u003e\n \u003cli\u003eL. F. a. W. I. Minha Lee, \u0026ldquo;Brokerbot: A Cryptocurrency Chatbot in the Social-technical Gap of Trust,\u0026rdquo; \u003cem\u003eComputer Supported Cooperative Work (CSCW),\u0026nbsp;\u003c/em\u003evol. 30, pp. 79-117, 2021.\u003c/li\u003e\n \u003cli\u003eD. H. D. L. Z. Z. \u0026amp;. N. C. Shaokang Cai, \u0026ldquo;An Reinforcement Learning-based Speach Censorship Chatbot System,\u0026rdquo; \u003cem\u003eThe Journal of Supercomputing ,\u0026nbsp;\u003c/em\u003evol. 78, pp. 8751-8773, 2022.\u003c/li\u003e\n \u003cli\u003eN. Haristiani, \u0026ldquo;Artificial Intelligence (AI) Chatbots as Language Learning Medium: An inquiry,\u0026rdquo; in \u003cem\u003eInternational Conference on Education, Science and Technology 2019\u003c/em\u003e, Gothenburg, Sweden, 2019.\u003c/li\u003e\n \u003cli\u003eL. B.-G. a. F. G. Jorge Cordero, \u0026ldquo;Use of Chatbots for Customer Serivces in MSMEs,\u0026rdquo; \u003cem\u003eApplied Computing and Informatics ,\u0026nbsp;\u003c/em\u003evol. 16, no. 1, 2022.\u003c/li\u003e\n \u003cli\u003ea. S. H. Darius Zumstein, \u0026ldquo;Chatbots-and Interactive Technology for Personalized Communication, Transactions and Services,\u0026rdquo; \u003cem\u003eIADIS International Journal on WWW/Internet,\u0026nbsp;\u003c/em\u003evol. 15, no. 1, pp. 96-109, 2018.\u003c/li\u003e\n \u003cli\u003eC. M. a. F. P. Jide Edu, \u0026ldquo;Exploring the Security and Privacy Risk of Chatbots in Messaging Services,\u0026rdquo; in \u003cem\u003eProceeding of the 22nd ACM Internet Measurement Conference (IMC\u0026apos;22)\u0026nbsp;\u003c/em\u003e, France, 2022.\u003c/li\u003e\n \u003cli\u003eF.-Y. L. a. J.-W. L. Sen-Tarng Lai, \u0026ldquo;A Banking Chatbot Security Control Procedure for Protecting User Data Security and Privacy,\u0026rdquo; in \u003cem\u003e13th International Conference on Broadband and Wireless Computing Communication and Applications (BWCCA-2018)\u003c/em\u003e, Switzerland, 2019.\u003c/li\u003e\n \u003cli\u003eL. P. \u0026amp;. A. D. V. Lucrezia Maria de Cosmo, \u0026ldquo;The role of attitude toward chatbots and privacy concern on the relationship between attitude toward mobile advertising and behavioral intent to use chatbots,\u0026rdquo; \u003cem\u003eItalian Jouranal of Marketing,\u0026nbsp;\u003c/em\u003epp. 83-102, 2021.\u003c/li\u003e\n \u003cli\u003e\u0026amp;. F. W. Josip Bozic, \u0026ldquo;Security Testing for Chatbots,\u0026rdquo; in \u003cem\u003eIn IFIP interanation Conference on Testing Software and System (ICTSS)\u003c/em\u003e, Switzerland, 2018.\u003c/li\u003e\n \u003cli\u003eC. M. F. P. e. a. Jide Edu, \u0026ldquo;Exploring the Security and Privacy Risks of Chatbots in Messaging Services,\u0026rdquo; in \u003cem\u003eIn Proceedings of the 22nd ACM Internet Measurement Conference (IMC \u0026apos;22)\u0026nbsp;\u003c/em\u003e, New York, USA, 2022.\u003c/li\u003e\n \u003cli\u003eM. A. a. G. M. Koien, \u0026ldquo;Cyber Security and the Internet of Things: Vulnerabilities, Threats, Intruders and Attacks,\u0026rdquo; \u003cem\u003eJournal of Cyber Security,\u0026nbsp;\u003c/em\u003evol. 4, pp. 65-88, 2015.\u003c/li\u003e\n \u003cli\u003ea. R. H. Rohan Kar, \u0026ldquo;Applying Chatbots to the Internet of Things: Opportunities and Architectureal Elements,\u0026rdquo; \u003cem\u003eInternational Journal of Advanced Comptuer Science and Application (IJACSA),\u0026nbsp;\u003c/em\u003evol. 7, no. 11, 2018.\u003c/li\u003e\n \u003cli\u003eA. A. e. a. Avijit Mallik, \u0026ldquo;Man-in-the-middle Attack: Understanding in Simple words,\u0026rdquo; \u003cem\u003eInternational Jouranl of Data and Network Science,\u0026nbsp;\u003c/em\u003evol. 3, pp. 77-92, 2019.\u003c/li\u003e\n \u003cli\u003eD. S. e. a. Laura Feinstein, \u0026ldquo;Statistical Approaches to DDoS Attack Detections and Response,\u0026rdquo; in \u003cem\u003eProceedings of the DARPA Information Survivability Conference and Exposition\u0026nbsp;\u003c/em\u003e, 2013.\u003c/li\u003e\n \u003cli\u003eK. M. Y. N. a. Yameen Ajani, \u0026ldquo;Homomorphic encryption for Secure Conversation with AI bots over cloud to prevent Social Engineering attacks,\u0026rdquo; \u003cem\u003eInternational Journal of Engineering Research and Applications,\u0026nbsp;\u003c/em\u003epp. 21-27, 2021.\u003c/li\u003e\n \u003cli\u003eT. V. Pathrabe, \u0026ldquo;Survey on Security Issues of Growing Technology: Big Data,\u0026rdquo; in \u003cem\u003eNational Conference on Latest Trends in Networking and Cyber Security\u003c/em\u003e, 2017.\u003c/li\u003e\n \u003cli\u003eS. Jung, \u0026ldquo;Semantic Vector Learning for Natural Language Understanding,\u0026rdquo; \u003cem\u003eScience Direct,\u0026nbsp;\u003c/em\u003evol. 56, pp. 130-145, 2019.\u003c/li\u003e\n \u003cli\u003eS. F. e. a. Eric Wallace, \u0026ldquo;Universal Adversarial Triggers for Attacking and Analyzing NLP,\u0026rdquo; in \u003cem\u003eProceedings of the 2019 Conferene on Empirical Methods in natural Language Processing\u0026nbsp;\u003c/em\u003e, Hong Kong, China, 2019.\u003c/li\u003e\n \u003cli\u003eB. W. R. G. e. a. Jiwei Tian, \u0026ldquo;Advarsarial Attacks and Defenses for Deep-Learning-Based Unmanned Aerial Vehicles,\u0026rdquo; \u003cem\u003eIEEE Internet of Things Journal,\u0026nbsp;\u003c/em\u003evol. 9, no. 22, pp. 22399-22409, 2022.\u003c/li\u003e\n \u003cli\u003eB. W. ,. Z. W. e. a. Jiwei Tiam, \u0026ldquo;Joint Adversarial Example and False Data Injections Attacks for State Estimation in Power Systems,\u0026rdquo; \u003cem\u003eIEEE Transactions on Cybernetics,\u0026nbsp;\u003c/em\u003evol. 52, no. 12, pp. 13699-13713, 2022.\u003c/li\u003e\n \u003cli\u003eT. K. D. a. C. P. Fran Casino, \u0026ldquo;A systematic literature review of blockchain-based applications: current status, classification and open issues,\u0026rdquo; \u003cem\u003eTelematics and Informatics,\u0026nbsp;\u003c/em\u003evol. 36, pp. 55-81, 2019.\u003c/li\u003e\n \u003cli\u003eY.-L. C. L. Y. P. \u0026amp;. C. S. K. Jing Yang, \u0026ldquo;A systematic Literature Review of Information Security in Chatbots,\u0026rdquo; \u003cem\u003eApplied Sciences,\u0026nbsp;\u003c/em\u003evol. 13, pp. 2-18, 2023.\u003c/li\u003e\n \u003cli\u003eM. W. F. a. G. I. S. Michael Lahzi Gaid, \u0026ldquo;Secure Translation Using Fully Homomorphic Encryption and Sequence-to-Sequence Neural Network,\u0026rdquo; in \u003cem\u003e28th International Conference on Computer Theory and Application (ICCTA)\u003c/em\u003e, Alexandria, Egypt, 2018.\u003c/li\u003e\n \u003cli\u003ea. H. C. Nasir Ahmad Jalali, \u0026ldquo;Federated Learning Security and Privacy-Preserving Algorithm and Experimental Research Under Internet of Things Critical Infrastracture,\u0026rdquo; \u003cem\u003eTsinghua Science and Technology ,\u0026nbsp;\u003c/em\u003evol. 29, no. 2, pp. 400-414, 2023.\u003c/li\u003e\n \u003cli\u003eH. A. a. A. U. Abbas Acar, \u0026ldquo;A Survey on Homomorphic Encryption Scheme: Theory and Implementation,\u0026rdquo; \u003cem\u003eACM Computing Surveys,\u0026nbsp;\u003c/em\u003evol. 51, no. 4, pp. 79:1-79:35, 2018.\u003c/li\u003e\n \u003cli\u003eD. Romanov, \u0026ldquo;Secure Multi-party computation for supply chain collaboration,\u0026rdquo; Delft University of Technology, 2021.\u003c/li\u003e\n \u003cli\u003eG. Sebastian, \u0026ldquo;Privacy and Data Protection in ChatGPT and Other AI Chatbots: Strategies for Securing User Information,\u0026rdquo; Georgia Institiute of Technology, 2022.\u003c/li\u003e\n \u003cli\u003eN. A. J. \u0026amp;. H. Chen, \u0026ldquo;Security Issues and Solutions in Federated Learning Under IoT Critical Infrastructure,\u0026rdquo; \u003cem\u003eWireless Personal Communication,\u0026nbsp;\u003c/em\u003evol. 129, pp. 475-500, 2023.\u003c/li\u003e\n \u003cli\u003eH. M. B. A. e. a. Peter Kairouz, \u0026ldquo;Advanced and Open Problems in Federated Learning,\u0026rdquo; \u003cem\u003eArxive: Machine Learning,\u0026nbsp;\u003c/em\u003evol. 14, no. 1-2, 2021.\u003c/li\u003e\n \u003cli\u003eW. Z. Q. L. e. a. Renhao Lu, \u0026ldquo;Adaptive Asynchronous Federated Learning,\u0026rdquo; \u003cem\u003eFuture Generation Computer Systems,\u0026nbsp;\u003c/em\u003evol. 152, pp. 193-206, 2023.\u003c/li\u003e\n \u003cli\u003e\u0026amp;. Q. L. Winson Ye, \u0026ldquo;Chatbot Security and Privacy in the Age of Personal Assistants,\u0026rdquo; in \u003cem\u003e2020 IEEE/ACM Symoposium on Edge Computing (SEC)\u003c/em\u003e, 2020.\u003c/li\u003e\n \u003cli\u003eI. I. M. A. S. a. A. O. Peter Voege, \u0026ldquo;Smart Chatbot for User Authentication,\u0026rdquo; \u003cem\u003eElectronics,\u0026nbsp;\u003c/em\u003evol. 11, 2022.\u003c/li\u003e\n \u003cli\u003eS. N. \u0026amp;. M. K. Aneesa M P, \u0026ldquo;Face Recognition using CNN: A Systematic Review,\u0026rdquo; \u003cem\u003eInternational Journal of Engineering Research \u0026amp; Technology (IJERT),\u0026nbsp;\u003c/em\u003evol. 11, no. 6, pp. 182-185, 2022.\u003c/li\u003e\n \u003cli\u003eR. E. Endeley, \u0026ldquo;End-to-End Encryption in Messaging Services and National Security-Case of WhatsApp Messanger,\u0026rdquo; \u003cem\u003eJournal of Information Security,\u0026nbsp;\u003c/em\u003evol. 9, pp. 95-99, 2018.\u003c/li\u003e\n \u003cli\u003eG. Gentry, \u0026ldquo;A Fully Homomorphic Encryption Scheme,\u0026rdquo; Stanford University , California, 2009.\u003c/li\u003e\n \u003cli\u003eA. G. e. a. Marcelo Blatt, \u0026ldquo;Secure Large-scale Genome-wide Association Studies Using Homomorphic Encryption,\u0026rdquo; \u003cem\u003eProceedings of the National Academy of Sciences of the United State of America (PNAS),\u0026nbsp;\u003c/em\u003evol. 117, no. 21, pp. 11608-11613, 2020.\u003c/li\u003e\n \u003cli\u003eS. ,. M. A. Jamal N. Bani Salameh, \u0026ldquo;Evaluation of Cloud Computing Platform for Image Processing Algorithm,\u0026rdquo; \u003cem\u003eJournal of Engineering Science and Technology,\u0026nbsp;\u003c/em\u003evol. 14, no. 4, 2019.\u003c/li\u003e\n \u003cli\u003eK. M. e. a. Yammen Ajani, \u0026ldquo;Homomorphic Encryption Technology for Securing Data in Cloud Computing: A Survey,\u0026rdquo; \u003cem\u003eInternational Journal of Computer Application,\u0026nbsp;\u003c/em\u003evol. 160, no. 6, 2017.\u003c/li\u003e\n \u003cli\u003eL. T. Y. \u0026amp;. J. H. P. Sushil Kumar Singh, \u0026ldquo;FusionFedBlock: Fusion of Blockchain and Federated Learning to preserve privacy in industry 5.0,\u0026rdquo; \u003cem\u003eInforamtion Fusion,\u0026nbsp;\u003c/em\u003evol. 90, pp. 233-240, 2023.\u003c/li\u003e\n \u003cli\u003eG. B. G. F. e. a. Daniele Mazzei, \u0026ldquo;A Blockchain Tokenizer for Industrial IoT Trustless Application,\u0026rdquo; \u003cem\u003eFuture Generation Computer Systems,\u0026nbsp;\u003c/em\u003evol. 105, pp. 432-445, 2020.\u003c/li\u003e\n \u003cli\u003eY. Q. Y. X. a. L. G. Chenhao Xu, \u0026ldquo;Asynchronous Federated Learning on Heterogeneous Devices: A Survey,\u0026rdquo; \u003cem\u003eComptuer Science Review ,\u0026nbsp;\u003c/em\u003evol. 50, 2023.\u003c/li\u003e\n \u003cli\u003eF. C. \u0026amp;. M. O. A. Md Sadek Ferdous, \u0026ldquo;In Search of Self-Sovereign Identity leveraging Blockchain Technology,\u0026rdquo; \u003cem\u003eIEEE Access,\u0026nbsp;\u003c/em\u003evol. 7, pp. 103059-103079, 2019.\u003c/li\u003e\n \u003cli\u003e\u0026amp;. M. N. Abylay Satybaldy, \u0026ldquo;Reveiw of Technical for Privacy-Preserving Blockchain Systems,\u0026rdquo; in \u003cem\u003eProceeding of the 2nd ACM International Symposium on Blockchain and Secure Critical Infrastructure\u0026nbsp;\u003c/em\u003e, Taipei, Taiwan, 2020.\u003c/li\u003e\n \u003cli\u003e,. H. C. e. a. Pranab Kumar Bharimalla, \u0026ldquo;A Blokchain and NLP Based Electroni Health Record System: Indian subcontinent Context,\u0026rdquo; \u003cem\u003eInformatica ,\u0026nbsp;\u003c/em\u003evol. 45, pp. 605-616, 2021.\u003c/li\u003e\n \u003cli\u003eY. Z. J. Y. X. L. H. L. Kun Shao, \u0026ldquo;The Triggers that open the NLP model backdoors are hidden in th advaresial Samples,\u0026rdquo; \u003cem\u003eComptuer \u0026amp; Security ,\u0026nbsp;\u003c/em\u003evol. 118, 2022.\u003c/li\u003e\n \u003cli\u003eD. G. \u0026amp;. T. K. Ramanpreet Kaur, \u0026ldquo;Artificial Intelligence for Cybersecurity: Literature Review and Future Research Direction,\u0026rdquo; \u003cem\u003eInformation Fusion ,\u0026nbsp;\u003c/em\u003evol. 97, 2023.\u003c/li\u003e\n \u003cli\u003eO. F. F. T. I. a. I. A. O. Femi Emmanuel Ayo, \u0026ldquo;Machine Learning Techniques for Hate Speech Classification of Twitter Data: State-of-art, Future Challenges and Research Direction,\u0026rdquo; \u003cem\u003eComputer Science Review ,\u0026nbsp;\u003c/em\u003evol. 38, 2020.\u003c/li\u003e\n \u003cli\u003eH. Z. j. M. e. a. Junfei Yi, \u0026ldquo;A review on the Covid-19 Pandemic prevention and control system based on AI,\u0026rdquo; \u003cem\u003eEngineering Application of Artificial Intelligence ,\u0026nbsp;\u003c/em\u003evol. 114, 2022.\u003c/li\u003e\n \u003cli\u003eE. U. U. R. L. R. e. a. Ogobuchi Daniel Okey, \u0026ldquo;Investigating ChatGPT and cybersecurity: A Perspectived on topic modeling and Sentiment analysis,\u0026rdquo; \u003cem\u003eComptuers \u0026amp; Security,\u0026nbsp;\u003c/em\u003evol. 135, 2023.\u003c/li\u003e\n \u003cli\u003eC. Chaka, \u0026ldquo;Geerative AI Chatbots-ChatGPT versus YouChat versus Chatsonic: Use Case of Selected Areas of Applied English Language Studies,\u0026rdquo; \u003cem\u003eInternational Journal of Learning,Teaching and Eduactioanl Research ,\u0026nbsp;\u003c/em\u003evol. 22, no. 6, pp. 1-19, 2023.\u003c/li\u003e\n \u003cli\u003eA. A. a. Y. J. Muna AI-Hawawreh, \u0026ldquo;Chatgpt for Cybersecurity: Practical Applications, Challenges and future Directions,\u0026rdquo; \u003cem\u003eCluster Computing ,\u0026nbsp;\u003c/em\u003evol. 26, pp. 3421-3436, 2023.\u003c/li\u003e\n \u003cli\u003eI. C. O. K. e. a. Jan Kocon, \u0026ldquo;ChatGPT : Jack of all trades, master of none,\u0026rdquo; \u003cem\u003eInformation Fusion,\u0026nbsp;\u003c/em\u003evol. 99, 2023.\u0026nbsp;\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"cluster-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Cluster Computing](https://www.springer.com/journal/10586)","snPcode":"10586","submissionUrl":"https://submission.nature.com/new-submission/10586/3","title":"Cluster Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Chatbot, Security/Privacy, Federated learning, Full Homomorphic Encryption, Blockchain","lastPublishedDoi":"10.21203/rs.3.rs-3862540/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3862540/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eChatbot is an artificial intelligence application that can provide a conversational environment between human and machine. Most organizations and industries are willing to lay out their services through chatbot because it can provide 24/7 customer support. Meanwhile it raises security and privacy challenges like access control, data leakage during transmission, SQL injection attack, language model attack which make the users concerned about their data, performance and accuracy. Therefore this research paper proposed a comprehensive framework integrated blockchain, federated learning and fully homomorphic encryption algorithm with face recognition to solve above mentioned chatbot\u0026rsquo;s challenges. The experimental result shows that distributed system improves chatbot accuracy (90%) and more transaction in less time with more clients do not affect the performance. In contrast, more iteration and clients will decrease the accuracy, performance and transactions in centralized system. In addition, fully homomorphic encryption improve and speed up data encryption process. It encrypted more data (1792 MB) in a small amount of 1240 time/sec and conversation/transactions can be transferred via a secure network to ensure the confidentiality, integrity and authenticity of users\u0026rsquo; data. The implementation of such comprehensive framework in real-life can improve chatbot security that is actively work as a customer agent in organization.\u003c/p\u003e","manuscriptTitle":"Comprehensive Framework for Implementing Blockchain-enabled Federated Learning and Full Homomorphic Encryption for Chatbot security System","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-17 09:50:44","doi":"10.21203/rs.3.rs-3862540/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-03-17T22:08:06+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-03-03T15:35:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"c48d23a5-89df-4d7c-b036-bb2d476c2003","date":"2024-03-03T10:58:11+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-01-23T08:12:01+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"e984dcd3-699a-458a-85e0-626b363b9211","date":"2024-01-17T10:16:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"390d5a71-4e60-4f52-86e3-1768ccaae6cb","date":"2024-01-17T09:17:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"67155bed-8a56-4b1a-8709-425f4027ad2c_SNPRID","date":"2024-01-17T08:25:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"e3bedb3d-e46a-46b2-a252-497c47c7a7db","date":"2024-01-17T08:06:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"97fa0c20-8f32-450c-8711-d07fae5395d5","date":"2024-01-17T07:51:40+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-01-17T07:49:20+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-01-16T12:30:02+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-01-16T07:39:25+00:00","index":"","fulltext":""},{"type":"submitted","content":"Cluster Computing","date":"2024-01-14T07:24:20+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"cluster-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Cluster Computing](https://www.springer.com/journal/10586)","snPcode":"10586","submissionUrl":"https://submission.nature.com/new-submission/10586/3","title":"Cluster Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"56919ee8-886e-4150-9b2c-c5b4efb3d91a","owner":[],"postedDate":"January 17th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2024-04-16T00:02:51+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-17 09:50:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3862540","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3862540","identity":"rs-3862540","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}