Privacy Risks Research Articles

The Skewed Privacy Concerns of Bystanders in Smart Environments

As ubiquitous computing brings sensors and actuators directly into our homes, they introduce privacy concerns for the owners and bystanders. However, privacy concerns may vary among devices and depend on the bystanders' social relation to the owner. In this work, we hypothesize 1) that bystanders assign more privacy concerns to smart home devices than personal computing devices, such as smartphones, even though they have the same capabilities, and 2) that a stronger social relationship mitigates some of the bystanders' privacy concerns. By conducting an online survey (n=170), we found that personal computing devices are perceived as significantly less privacy-concerning than smart home devices while having equal capabilities. By varying the assumed social relationship, we further found that a stronger connection to the owner reduces privacy concerns. Thus, as bystanders underestimate the risk of personal computing devices and are generally concerned about smart home devices, it is essential to alert the user about the presence of both. We argue that bystanders have to be informed about the privacy risks while entering a new space, in the best case, already in the entrance area.

Support Personalized Weighted Local Differential Privacy Skyline Query

The potential privacy risks in certain situations are of concern because of the frequent sharing of data during skyline queries, leading to leakage of users’ private information. The most common privacy-preserving technique is to anonymize data by removing or changing certain information, for which an attack with specific background knowledge would render the privacy protection ineffective. To overcome these difficulties, this study proposes a personalized weighted local differential privacy method (PWLDP) to protect data privacy during skyline querying. Compared with existing studies of skyline queries under privacy protection, the degree of privacy protection can be quantitatively analyzed, and the processing of data privacy lies with the user, who quantitatively perturbs the processing according to the sensitivity of the weights of different attributes to avoid substantial information loss. The performance of the proposed PWLDP is verified by comparing PWLDP and LDP on different datasets, the average privacy leakage reduction of 62.22% and 51.67% is obtained for experiments conducted on different datasets relative to the iDP-SC algorithm, and the experimental results demonstrate the efficiency and advantages of the proposed method.

Privacy concerns can stress you out: Investigating the reciprocal relationship between mobile social media privacy concerns and perceived stress

Abstract Mobile social media have become a widespread means to participate in everyday social and professional life. These platforms encourage the disclosure and exchange of personal information, which comes with privacy risks. While past scholarship has listed various predictors and consequences of online privacy concerns, there has been to date no empirical investigation of a conceivable relationship with perceived stress. Using a longitudinal panel study, we examined the reciprocal relationship between mobile social media privacy concerns and perceived stress. Results supported the hypothesis that mobile social media privacy concerns at T1 are associated with higher perceived stress at T2. However, we found no evidence for the reverse association, that is, perceived stress at T1 was not related to mobile social media privacy concerns at T2. The findings are discussed based on two models—the “Antecedents privacy concerns outcomes” model and the “Privacy calculus” model.

Reasons for qualitative psychologists to share human data.

Qualitative data sharing practices in psychology have not developed as rapidly as those in parallel quantitative domains. This is often explained by numerous epistemological, ethical and pragmatic issues concerning qualitative data types. In this article, I provide an alternative to the frequently expressed, often reasonable, concerns regarding the sharing of qualitative human data by highlighting three advantages of qualitative data sharing. I argue that sharing qualitative human data is not by default 'less ethical', 'riskier' and 'impractical' compared with quantitative data sharing, but in some cases more ethical, less risky and easier to manage for sharing because (1) informed consent can be discussed, negotiated and validated; (2) the shared data can be curated by special means; and (3) the privacy risks are mainly local instead of global. I hope this alternative perspective further encourages qualitative psychologists to share their data when it is epistemologically, ethically and pragmatically possible.

Privacy, accuracy, and model fairness trade-offs in federated learning

As applications of machine learning become increasingly widespread, the need to ensure model accuracy and fairness while protecting the privacy of user data becomes more pronounced. On this note, this paper introduces a federated learning training model, which allow clients to simultaneously learn their model and update the associated parameters on a centralized server. In our approach, we seek to achieve an acceptable trade-off between privacy, accuracy, and model fairness by using differential privacy (DP), which also helps to minimize privacy risks by protecting the presence of a specific sample in the training data. Machine learning models can, however, exhibit unintended behaviors, such as unfairness, which result in groups with certain sensitive characteristics (e.g., gender) receiving different patterns of outcomes. Hence, we discuss the fairness and privacy effect of local DP and global DP when applied to federated learning by designing a fair and privacy quantification mechanism. In doing so, we can achieve an acceptable trade-off between accuracy, privacy, and model fairness. We quantify the level of fairness based on the constraints of three definitions of fairness, including demographic parity, equal odds, and equality of opportunity. Finally, findings from our extensive experiments conducted on three real-world datasets with class imbalance demonstrate the positive effect of local and global DP on fairness. Our study also shows that privacy can come at the cost of fairness, as stricter privacy can intensify discrimination. Hence, we posit that careful parameter selection can potentially help achieve a more effective trade-off between utility, bias, and privacy.

Addressing privacy risk in neuroscience data: from data protection to harm prevention.

A recent increase in the amount and availability of neuroscience data within and outside of research and clinical contexts will enhance reproducibility of neuroscience research leading to new discoveries on the mechanisms of brain function in healthy and disease states. However, the uniquely sensitive nature of neuroscience data raises critical concerns regarding data privacy. In response to these concerns, various policy and regulatory approaches have been proposed to control access to and disclosure of neuroscience data, but excessive restriction may hamper open science practice in the field. This article argues that it may now be time to expand the scope of regulatory discourse beyond protection of neuroscience data and to begin contemplating how to prevent potential harm. Legal prohibition of harmful use of neuroscience data could provide an ultimate safeguard against privacy risks and would help us chart a path toward protecting data subjects without unduly limiting the benefits of open science practice. Here we take the Genetic Information Non-Discrimination Act (GINA) as a reference for this new legislation and search for answers to the core regulatory questions based on what we have learned from the enactment of the GINA and the merits and weaknesses of the protection it provides.

Elucidating the role of emotion in privacy-concerns: A text-Convolutional Neural Network (Text-CNN)-based tweets analysis of contact tracing apps

The extant contact tracing privacy literature is yet to explore the significance of user emotions in privacy-related decision-making such as whether to use such potentially privacy-invasive apps. Using social media analytics, the present study examines users’ privacy-related emotions stimulated by privacy-related aspects of contact tracing apps. A text-Convolutional Neural Network (Text-CNN)-based emotion analysis of tweets on the Indian contact tracing app Aarogya Setu and its Singaporean counterpart TraceTogether conducted in the paper reveals that users expressed negative privacy-related emotions towards these apps indicating high levels of perceived privacy risks and the perceived lack of privacy protection. For TraceTogether, users have also exhibited positive emotions to appreciate the steps taken by the government to protect their privacy. Based on these findings, the government/data controllers can devise strategies to assuage users’ negative emotions and promote positive emotions to encourage the adoption of contact tracing apps. This work incorporates privacy related emotions as key informants about user privacy concerns within the Privacy Calculus Theory. By relying on candid user opinions available through rich but inexpensive user-generated content, the research provides a quick, reliable, and cost-effective approach to study potential app users’ emotions to gain insights into privacy concerns related to any e-governance platform.

Should the chatbot “save itself” or “be helped by others”? The influence of service recovery types on consumer perceptions of recovery satisfaction

The application of artificial intelligence is considered essential to adapt to a new cycle of industrial transformation and technological advancements in many fields and industries. The extensive use of artificial intelligence technology is expected to improve the level and quality of services provided by companies adopting these methods. In this study, we propose a novel approach to self-recovery by chatbot systems after service failures based on social response theory. Moreover, we explore differences in consumer perceptions of different service recovery types and their impact on recovery satisfaction, and discusses whether the intelligence of the computational agent also has an effect.We present the results of three scenario-based experiments, which demonstrate the positive effect of chatbot self-recovery on consumer satisfaction, and show the mediating paths of service recovery types in terms of perceived functional value and privacy risks as well as the boundary condition of the level of robot intelligence. This work expands the range of applications of chatbots in the service industry and provides a new framework for the governance of artificial intelligence.

XAuth: Efficient Privacy-Preserving Cross-Domain Authentication

It is well known that each Public Key Infrastructure (PKI) system forms a closed security domain and only recognizes certificates in its own domain (such as medical systems, financial systems, and 5G networks). When users need to access services in other domains, their identities often cannot be recognized or PKI systems require extremely complex operations to authenticate the users’ identities. This is the cross-domain authentication problem. The distributed consensus feature of blockchain provides a technical approach to solve this problem. However, there are some unresolved problems in existing blockchain-based schemes. On one hand, due to the low throughput of blockchain systems, the response speed may be insufferable when the number of cross-domain authentication requirements becomes enormous. On the other hand, these schemes insufficiently consider the privacy risk in the cross-domain scenario. In this article, we propose an efficient privacy-preserving cross-domain authentication scheme called XAuth that is integrated naturally with the existing PKI and Certificate Transparency (CT) systems. Specifically, we design a lightweight correctness verification protocol based on Multiple Merkle Hash Tree for rapid response. To protect users’ privacy, we present an anonymous authentication protocol for cross-domain authentication. The security analysis and experimental results demonstrate that XAuth is secure and efficient.

Motivating Factors to Self-Disclosure on Social Media: A Systematic Mapping

<bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Background:</b> Self-disclosure on social media can cause some privacy risks, but it benefits users and businesses if adequately managed. Companies may benefit from users’ self-disclosure on social media to better understand the consumers’ needs, customize services or products, and address users’ concerns to sculpt positive brand reputation, trust, and sales. In addition, users’ better understanding of self-disclosure motivations helps them manage more suitable topic, platform, and concepts to match the intended online personal or professional persona. Finally, technical communicators’ understanding of the motivation for social media self-disclosures can help them leverage available self-disclosure in producing more effective technical communication and carefully plan self-disclosures with clear motivations. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Literature review:</b> Many researchers have studied the varying self-disclosure motivations, but to the best of our knowledge, no mapping studies are currently available summarizing the motivations of self-disclosure on social media. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Research question:</b> What does the current research about self-disclosure identify as users’ motivation for self-disclosure on social media? <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Research methodology:</b> We conducted a systematic mapping study that included relevant journal and conference publications. Mapping studies are suitable for structuring a broad research field concerning research questions about content, methods, or trends in the existing publications. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Results:</b> We found four categories for social media self-disclosures: discloser-related, audience-related, platform- and affordances-related, and perceived risk- and cost-related. Within the main categories, we found varying submotivations. We also discuss the implications of our findings and future research needs. <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Conclusions:</b> The mapping of available studies helps researchers, academics, and practitioners understand existing self-disclosure motivations and research gaps. In addition, social media stakeholders planning to use social media self-disclosures within their areas of interest can use this study as a starting point to understand what drives social media self-disclosures.

Privacy and Security of Telehealth Services

Telehealth technologies have aided in the distribution of health-related services, especially during the COVID-19 pandemic. With increased telehealth use, privacy risks and security concerns among healthcare providers and patients have subsequently increased. In previous research where 205 research papers were reviewed related to healthcare privacy and security, the results reveal a significant lack of research on telehealth security and privacy. In this panel, we aim to discuss this critical gap by studying how current healthcare process designs and provider workflows associated with the use of telehealth technologies impact the security and privacy of patients' protected health information. The panel will engage in in-depth discussions about value-sensitive design while focusing on potentially critical data, such as healthcare data which is not typically emphasized in general security and privacy concerns. To address this, the panel consists of experts from the diverse field including privacy and security, user interaction, and healthcare.

Security and Privacy in Artificial Intelligence-Enabled 6G

The sixth-generation (6G) mobile communication network is expected to provide world-connected smart and autonomous services by leveraging artificial intelligence (AI) technologies. In particular, federated learning (FL) is advocated to promote the implementations of ubiquitous AI applications in 6G, where massive distributed data is utilized to cooperatively train AI models as well as protect privacy and reduce resource consumption. However, due to the large number of interactions among mobile devices or infrastructures in FL, the AI inevitably suffers from a series of potential privacy and security risks. In this article, we investigate the security and privacy of AI-enabled 6G. Specifically, we first present the AI-enabled 6G architecture, including the space-air-ground-ocean integrated network (SAGOIN) architecture and two different FL-based AI model training frameworks in 6G. Furthermore, we discuss in detail the privacy and security threats, including cheating attack, low-quality local model training attack, and privacy stealing attack. Afterwards, we present a case study to design the security and privacy preservation scheme by integrating the trust evaluation, Q-learning based incentive, and local differential privacy (LDP) mechanism. Finally, we give out several future research directions in promising AI-enabled 6G.

Data Privacy and Trustworthy Machine Learning

The privacy risks of machine learning models is a major concern when training them on sensitive and personal data. We discuss the tradeoffs between data privacy and the remaining goals of trustworthy machine learning (notably, fairness, robustness, and explainability).

A digital mask to safeguard patient privacy

The storage of facial images in medical records poses privacy risks due to the sensitive nature of the personal biometric information that can be extracted from such images. To minimize these risks, we developed a new technology, called the digital mask (DM), which is based on three-dimensional reconstruction and deep-learning algorithms to irreversibly erase identifiable features, while retaining disease-relevant features needed for diagnosis. In a prospective clinical study to evaluate the technology for diagnosis of ocular conditions, we found very high diagnostic consistency between the use of original and reconstructed facial videos (κ ≥ 0.845 for strabismus, ptosis and nystagmus, and κ = 0.801 for thyroid-associated orbitopathy) and comparable diagnostic accuracy (P ≥ 0.131 for all ocular conditions tested) was observed. Identity removal validation using multiple-choice questions showed that compared to image cropping, the DM could much more effectively remove identity attributes from facial images. We further confirmed the ability of the DM to evade recognition systems using artificial intelligence-powered re-identification algorithms. Moreover, use of the DM increased the willingness of patients with ocular conditions to provide their facial images as health information during medical treatment. These results indicate the potential of the DM algorithm to protect the privacy of patients’ facial images in an era of rapid adoption of digital health technologies.

Private record linkage with linkage maps

AbstractPrivate record linkage is an actively pursued research area to facilitate the linkage of database records under the constraints of regulations that do not allow linkage agents to learn sensitive identities of record owners. Recent works have shown that linkage using commutative ciphers, which were discarded earlier for efficiency concerns, can be made feasible by leveraging precomputations, data parallelism, and probabilistic key reuse approaches. In this work, we propose further optimizations that can be performed to improve the runtime efficiency of such an approach. We transition from modular exponentiation ciphers to elliptic curve operations to improve precomputation time, eliminate memory intensive comparisons of encrypted values, and introduce data structures to detect negative comparisons. We benchmark the proposed approach using real world demographics data, and provide an extensive study of the parametric aspects of the approach. We also supplement our execution time results with an assessment of the residual privacy risk left by the approach. The approach can perform a linkage of two datasets with 105 records each in 20 minutes in a commodity laptop. This is achieved by eliminating the need to compare more than 70% of the record pairs. By design, the linkage accuracy is also retained at the same level as a nonprivate record linkage procedure.

The Determinants of User Acceptance of Mobile Medical Platforms: An Investigation Integrating the TPB, TAM, and Patient-Centered Factors.

Mobile medical platforms (MMPs) can make medical services more accessible and effective. However, the patient-centered factors that influence patients’ acceptance of MMPs are not well understood. Our study examined the factors affecting patients’ acceptance of MMPs by integrating the theory of planned behavior (TPB), the technology acceptance model (TAM), and three patient-centered factors (i.e., perceived convenience, perceived credibility, and perceived privacy risk). Three hundred and eighty-nine Chinese respondents were recruited in this study and completed a self-administered online questionnaire that included items adapted from validated measurement scales. The partial least squares structural equation modeling results revealed that perceived privacy risk, perceived credibility, and perceived ease of use directly determined the perceived usefulness of an MMP. Perceived convenience, perceived credibility, and perceived usefulness significantly affected the patients’ attitudes toward MMPs. Perceived usefulness, attitude, perceived privacy risk, and perceived behavioral control were important determinants of the patients’ behavioral intentions to use MMPs. Behavioral intention and perceived behavioral control significantly influenced perceived effective use. Perceived credibility and perceived ease of use significantly affected perceived convenience. However, social influence had no significant effect on attitude and behavioral intention. The study provides important theoretical and practical implications, which could help practitioners enhance the patients’ use of MMPs for their healthcare activities.

Federated Learning for cross-jurisdictional analyses: A case study.

The objective of this project is to implement a harmonized artificial intelligence (AI)-based de-identification of free-text medical data across multiple Canadian jurisdictions. This federated learning approach will allow these jurisdictions to leverage each other’s data and resources while no individual-level data leaves the jurisdiction.&#x0D; Federated Learning enables health data centers in different jurisdictions to collaborate in training machine learning models without sharing individual-level data. This approach will significantly reduce privacy and cybersecurity risks and barriers that are involved in sharing and moving data across different jurisdictions.&#x0D; In a federated learning environment, machine learning models are trained on multiple data sources available in local data centers; local data are not shared to a central computing/analysis environment. Instead, parameters (such as model weights) are shared between these local data centers to generate a global model that will be shared and used by all participating data centers.&#x0D; In this case study, four health research data centers in different Canadian provinces will take part in deployment of an AI-based application for de-identification of free-text data. The data centers are members of Health Data Research Network (HDRN) Canada. The deployment will include:&#x0D; &#x0D; harmonized annotation and labeling of local data,&#x0D; local training of entity recognition algorithms,&#x0D; integrating model weights from each data centers to create a global model&#x0D; development of license agreements between the participating data centers to allow sharing model weights&#x0D; &#x0D; This is an ongoing project. The talk will demonstrate learning experiences, advantages, and challenges in a federated learning environment and explore the feasibility of transporting this approach to other multi-jurisdiction research networks.

Ethical considerations in the use of Machine Learning for research and statistics.

This paper, based upon new guidance created in collaboration with researchers from several national statistical institutes, explores the main ethical considerations associated with the use of machine learning techniques for aggregate statistics. The aim of this paper is to provide applied, practical ethical guidance for researchers using machine learning techniques. Following an extensive literature review, alongside discussion and collaboration with a number of national statistical institutes, it was identified that there was a need for applied guidance on the use of machine learning for the production of official statistics by the international research and statistical community. Feedback was gathered from interested stakeholders, which found that whilst there were resources available to researchers relating to the ethical considerations of machine learning projects, these focus mainly on operational uses of machine learning, and furthermore, lacked advice on how to practically mitigate ethical issues that arise throughout the project lifecycle. The guidance focuses on four main ethical considerations, found to be prevalent within machine learning research, and offers ways to mitigate these issues should they arise. These are: the importance of minimising and mitigating social bias and discrimination within machine learning research, and clearly communicating these, and the limitations of our research; the need to consider the transparency and explainability of machine learning research, and the implications this has for reproducibility; the importance of maintaining accountability throughout machine learning processes, ensuring that models are used only for their intended purposes, and that different stakeholders are aware of their responsibilities; the need to consider the confidentiality and privacy risks arising from the data used, both in relation to training data which is fed into the machine, and outputs resulting from the machine learning’s findings. The guidance has been well-received following its release, and feedback from the wider user community to date has been positive. The IPDLN conference provides an opportunity for further feedback to be collated to ensure that the guidance continues to be valuable to its intended audience.

A new data linkage infrastructure.

Our overall objectives are to make data linkage a more feasible and frequently used approach by easing the burden on data controllers posed by data assembly for complex linkage projects. By moving towards data linkage as a more expected and standard approach, policy relevant research into a variety of outcomes can be conducted.&#x0D; The approach to this has been to create a new secure data linkage infrastructure, and to formulate research projects which will use the linked research data delivered by the infrastructure. A suite of information governance documents have also been developed for the data to be held in the secure infrastructure. Holding the data in a de-identified format on behalf of the data controller in a dedicated area for linkage projects means that the data controller only needs to be involved in the approvals process for the project, and a team of experts who operate the infrastructure can assemble the data.&#x0D; The results have been that there are several datasets pertaining to Education in Scotland being held within the infrastructure, with more in the pipeline. The data is held in such a way that there are no personal identifiers present in the same location as the variables which are typically of research interest; the stored data is split into three parts and shared across two partner organisations. This separation of functions approach assures data controllers and the public that the data poses extremely minimal privacy risk while in storage, but can be assembled into products which are very useful for approved research in the public interest.&#x0D; In conclusion, the new methodology for data linkage developed by this approach is hoped to save time, money and lives. A dedicated secure environment for data linkage will ensure the security and privacy of stored data while providing capacity for complex, rich research data to be brought together.

Brainprints: identifying individuals from magnetoencephalograms

Magnetoencephalography (MEG) is used to study a wide variety of cognitive processes. Increasingly, researchers are adopting principles of open science and releasing their MEG data. While essential for reproducibility, sharing MEG data has unforeseen privacy risks. Individual differences may make a participant identifiable from their anonymized recordings. However, our ability to identify individuals based on these individual differences has not yet been assessed. Here, we propose interpretable MEG features to characterize individual difference. We term these features brainprints (brain fingerprints). We show through several datasets that brainprints accurately identify individuals across days, tasks, and even between MEG and Electroencephalography (EEG). Furthermore, we identify consistent brainprint components that are important for identification. We study the dependence of identifiability on the amount of data available. We also relate identifiability to the level of preprocessing and the experimental task. Our findings reveal specific aspects of individual variability in MEG. They also raise concerns about unregulated sharing of brain data, even if anonymized.