Information Privacy Research Articles

The Impact of AI-Personalized Recommendations on Clicking Intentions: Evidence from Chinese E-Commerce

AI-personalized recommendation technology offers more accurate and diverse choices to consumers and increases click-through rates and sales on e-commerce platforms. Yet, data on consumers’ experiences of AI-personalized recommendations and their impact path on clicking intention are scarce. This article addressed these issues through three studies. In study 1, we adopted the Grounded Theory approach to conduct in-depth interviews with 30 Chinese consumers and constructed a scale to measure the impact of consumer experience on click intention. In study 2, we adopted the empirical research method to conduct reliability and validity tests on 347 valid questionnaires to finalize the scale officially. In study 3, based on the SOR theory, we constructed a model and formulated hypotheses and then conducted empirical analysis using 1097 valid questionnaires. We found that the relevance, inspiration, and insightful experiences of AI-personalized recommendations can significantly promote consumers’ clicking intention. Moreover, immersive experience mediates between the former three factors and clicking intention, and technology acceptance mediates between relevance, inspiration, and clicking intention. When consumers perceive a high degree of information privacy infringement, the immersive experience’s positive impact on clicking intention will be weakened. Meanwhile, the promoting effect of technology acceptance on clicking intention will also be inhibited. When information quality improves, the positive impact of technology acceptance on clicking intention will be enhanced. This research fills the gap in the literature on consumers’ experiences of AI-personalized recommendations and clarifies how these experiences affect the clicking intention. It offers valuable insights for e-commerce platforms to continuously optimize personalized recommendation algorithms and boost the click conversion rate of online shopping.

Enhancing User Differentiation in the Electronic Personal Synthesis Behavior (EPSBV01) Algorithm by Adopting the Time Series Analysis

The progress of contemporary technology has rendered information systems essential in our everyday existence, underscoring the crucial necessity to safeguard information security and privacy. In password authentication, the Electronic Personal Synthesis Behaviour (EPSB) heightens the accuracy of authorizing an authenticated user based on three parameters: EPSBERROR, EPSBTime, and EPSBStyle. EPSBTime suffers from a lack of indicators associated with the legitimate user; containing only six indicators, there arose the need to adopt methods for generating additional reliable indicators by analyzing old indicators and generating new indicators related to the legitimate user. Therefore, this study aims to test the impact of adopting time series analysis in the EPSB time indicator on improving the differentiation of user legitimacy in the case of password-stolen attacks. The research methodology, which involves analyzing and evaluating existing authentication methods in web-based systems, is a key component of this study. The study is divided into stages, with the first phase focusing on enhancing the existing EPSB model, the second phase implementing EPSBalgorithmV01, and the final stage ensuring validation. Thus, two preliminary experiments were conducted with 22 users from January 13 to February 1, 2024. The final phase involved comparing EPSBV01's accuracy in determining unauthorized users before and after using the ARIMA method. Thus, the EPSBV01algorithm successfully identified 17 unauthorized users during a stolen password attack simulation, outperforming the normal EPSB by 22.73%. Doi: 10.28991/ESJ-2025-09-01-014 Full Text: PDF

Identifying protected health information by transformers-based deep learning approach in Chinese medical text.

Purpose: In the context of Chinese clinical texts, this paper aims to propose a deep learning algorithm based on Bidirectional Encoder Representation from Transformers (BERT) to identify privacy information and to verify the feasibility of our method for privacy protection in the Chinese clinical context. Methods: We collected and double-annotated 33,017 discharge summaries from 151 medical institutions on a municipal regional health information platform, developed a BERT-based Bidirectional Long Short-Term Memory Model (BiLSTM) and Conditional Random Field (CRF) model, and tested the performance of privacy identification on the dataset. To explore the performance of different substructures of the neural network, we created five additional baseline models and evaluated the impact of different models on performance. Results: Based on the annotated data, the BERT model pre-trained with the medical corpus showed a significant performance improvement to the BiLSTM-CRF model with a micro-recall of 0.979 and an F1 value of 0.976, which indicates that the model has promising performance in identifying private information in Chinese clinical texts. Conclusions: The BERT-based BiLSTM-CRF model excels in identifying privacy information in Chinese clinical texts, and the application of this model is very effective in protecting patient privacy and facilitating data sharing.

Differentially private explanations for aggregate query answers

Differential privacy (DP) is the state-of-the-art and rigorous notion of privacy for answering aggregate database queries while preserving the privacy of sensitive information in the data. In today’s era of data analysis, however, it poses new challenges for users to understand the trends and anomalies observed in the query results: Is the unexpected answer due to the data itself, or is it due to the extra noise that must be added to preserve DP? In the second case, even the observation made by the users on query results may be wrong. In the first case, can we still mine interesting explanations from the sensitive data while protecting its privacy? To address these challenges, we present a three-phase framework DPXPlain, which is the first system to the best of our knowledge for explaining group-by aggregate query answers with DP. In its three phases, DPXPlain (a) answers a group-by aggregate query with DP, (b) allows users to compare aggregate values of two groups and with high probability assesses whether this comparison holds or is flipped by the DP noise, and (c) eventually provides an explanation table containing the approximately ‘top-k’ explanation predicates along with their relative influences and ranks in the form of confidence intervals, while guaranteeing DP in all steps. We perform an extensive experimental analysis of DPXPlain with multiple use-cases on real and synthetic data showing that DPXPlain efficiently provides insightful explanations with good accuracy and utility.

FRAUD DETECTION IN FINANCIAL TRANSACTONS

Financial fraud, considered as deceptive tactics for gaining financial benefits, has recently become a widespread menace in companies and organizations. Conventional techniques such as manual verifications and inspections are imprecise, costly, and time consuming for identifying such fraudulent activities. With the advent of artificial intelligence, machine-learning-based approaches can be used intelligently to detect fraudulent transactions by analyzing a large number of financial data. Fraud has been a persistent challenge in the realm of financial transactions, posing significant threats to businesses, financial institutions, and individuals alike. As technology advances and financial systems become increasingly digitalized, the methods and sophistication of fraudulent activities also evolve. This paper explores cutting edge technologies such as machine learning and are revolutionizing fraud detection. By analyzing transaction patterns and employing anomaly detection algorithms, such as CNN organizations can identify and mitigate fraudulent activities in real time. In response to this ongoing threat, the field of fraud detection in financial transactions has emerged as a critical area of focus for organizations worldwide. In order to identify fraudulent behavior in financial transactions, this research paper suggests a unique technique. Key challenges include handling imbalanced datasets where fraudulent transactions are rare compared to legitimate ones, ensuring the privacy and security of sensitive financial information, and maintaining low latency to prevent delays in transaction processing.

Applying the S-O-R Framework in Online Hotel Bookings: Literature Review

This study aims to investigate the factors influencing customers' online hotel booking intentions in the post-COVID-19 pandemic context. Drawing on the Stimulus-Organism-Response (S-O-R) framework, the research proposes to measure three stages leading to behavioral intention: the stimulus stage, the organism stage, and the response stage. The findings are expected to provide valuable insights for the hospitality industry to better understand and respond to changing customer behavior in the post-pandemic recovery period. Base on reviewed previous research about S-O-R model and hotel booking intention, this article proposition five independent variables including information quality, perceived interactivity, safety and privacy, price and promotion and electronic word of mount (e-WOM) as stimulus stage. These stimuli are possible to effect to perceived value in organism stage and positive push up Online hotel booking intention. As this research mainly focus on conceptual framework propose, this framework need to be test in more specific context to prove the practical applicable in hospitality industry.

Combining Differential Privacy with FedAvgM to Optimize Data Access Control in the Information Platform

In response to the privacy leakage risk and low-efficiency issues of traditional data access control, especially in distributed environments, data transmission and centralized processing may expose sensitive information. To improve data access control in information platforms, ensure data security, and accelerate global model training, this article introduced a joint mechanism of differential privacy and FedAvgM (Federated Averaging with Momentum) optimization to enhance privacy protection and improve training efficiency. Firstly, the Laplacian noise mechanism was adopted to prevent user privacy information from being leaked during global model training by adding noise to the dataset, ensuring the privacy of each data access. Then, combined with the FedAvgM optimization algorithm, distributed nodes calculated local model parameters separately and merged these parameters through weighted averaging to reduce the training time of the global model and improve efficiency. Finally, layered encryption technology was adopted to add multiple layers of encryption during data transmission to ensure the security of the transmission link. At the same time, a dynamic permission allocation mechanism was introduced to limit the frequency of access to sensitive data. The experimental results demonstrate that when the privacy budget at ε=20, the privacy protection method of Laplace mechanism still maintains 73% accuracy in data transmission, with a privacy leakage risk of only 0.27. Under the same training epochs, the FedAvgM optimization algorithm achieved a correct data transmission accuracy of 97%. In comparison, the convergence speed increased by 0.29%/min in 1-20 training epochs, showing a faster convergence speed. Under four different noise attack methods, the anti-noise ability of layered encryption is the lowest at 15.4 dB, 18.5 dB, 17.1 dB, and 17.3 dB, respectively. This method effectively improves the problems of privacy leakage risk and low efficiency in traditional data access control.

Behavioural intention to adopt mobile trading apps: an integrated theoretical and digital framework, privacy concerns, and information richness model

This research investigates the transformative impact of mobile trading apps on the Indian financial landscape, particularly in the context of the unprecedented surge in DMAT (dematerialisation) accounts following the COVID-19 pandemic. Due to the advancement of online platforms and fast internet connectivity, stock exchanges across the globe have seen a dramatic inflow of retail investors and brokerage firms. The choice of using a particular mobile trading app draws significant importance because various factors determine the ability and ease of use of a specific app. The study underscores the role of FinTech services, particularly mobile trading apps, in revolutionizing stock trading by offering real-time access, increased trading activity, and enhanced features. Despite the proliferation of research on FinTech apps, a notable gap exists in understanding the adoption dynamics of mobile trading apps, especially in the Indian context. To address this gap, our research applies an adapted and extended version of the Unified Theory of Acceptance and Use of Technology (UTAUT-3) framework to examine the factors influencing investors’ intentions and usage of mobile trading apps. We introduce novel elements such as information richness and privacy concerns, which are crucial in the financial domain. A convenient sample size of 573 actively brokerage app-using respondents was selected to investigate and conclude the consumers’ behavioural intention to use mobile trading apps. The findings highlight the significant impact of practical value, effort expectancy, social influence, hedonic motivation, trust, information richness, privacy concerns, facilitating conditions, and personal innovativeness in IT (Information technology) on investors’ intentions to use trading apps. These factors influence behavioural intentions and mediate the relationship between various constructs, emphasizing their multifaceted roles in shaping user perceptions. Theoretical implications of the research contribute to extending the UTAUT-3 model and providing a comprehensive framework for examining technology adoption in the financial domain. Moreover, practical implications guide developers, financial institutions, and policymakers in creating secure, user-friendly, and information-rich mobile trading systems. While acknowledging sampling and self-reported data limitations, this research lays the groundwork for future longitudinal studies. It encourages the exploration of diverse FinTech services to gain a holistic understanding of adoption dynamics in the evolving financial technology landscape. This study adds empirical knowledge to mobile trading app adoption and catalyzes further research, shaping the trajectory of FinTech studies and practical applications in the ever-evolving financial ecosystem.

'Stuck or unstable': partnerships between GPs and patients with complex chronic conditions. A qualitative study.

In chronic care, patient-GP collaboration is essential, but might be challenging if patients have complex health problems due to multimorbidity, psychosocial predicaments and addiction problems. To understand and manage these challenges, it is important to explore how patients' and GPs' attempt to collaborate, to maintain and achieve an alliance in order to gain good quality of care. To explore how dyads of GPs and patients that GPs deem have complex health problems and difficulties following treatment perceive and manage challenges in their chronic care partnership. A qualitative study from Danish general practice in deprived, rural areas. Semi-structured interviews with 12 dyads of GPs and patients with doctor-assessed complex chronic conditions and difficulties following treatment. The principles of Systematic Text Condensation were used in the analysis. Overall, the patient-GP collaboration could be characterized as either stuck or unstable. In both types, the challenges were identified as pointless consultations, conflicts about lifestyle, resignation, concealment of information, and hopelessness. These challenges could be managed by solving conflicts, adjusting to the patient's needs, accommodating the challenges in the relationship, and offering continued emotional support even with unsolved medical problems. Care of patients with complex health problems may possess several challenges. In this study, patients and GPs experienced the relational dimension as crucial for collaboration. A robust therapeutic alliance, incorporating the patient's agenda, offers an essential foundation for enhancing care in individuals with complex health problems.

Deontological aspects of doctor’s interaction with vulnerable patient groups

The article highlights the deontological aspects of the interaction between physicians and vulnerable patient groups as a key direction of modern ethical and legal regulation of medical practice. The importance of combining ethical principles and legal norms to foster trustful relationships, ensure quality and equitable medical care regardless of social status, age, or health condition of the patient is emphasized. Particular attention is given to informed consent, which guarantees respect for patient autonomy, and the protection of medical confidentiality, ensuring the privacy and security of personal information. The issue of professional ethics for physicians is examined in the context of preventing discriminatory practices and upholding patient rights, especially for those requiring enhanced protection. The legal regulation of medical activities is presented by the author as a foundation for improving service efficiency and strengthening public trust in healthcare. The author concludes that professional associations and ethical committees play a crucial role in improving medical deontology: they develop standards, monitor compliance, and provide consultations, contributing to the dynamic development of professional culture. The article underscores the need for specialized educational programs for medical professionals to cultivate a profound understanding of ethical and legal aspects, as well as consideration of patients’ cultural, social, and economic characteristics. This is particularly relevant in the context of implementing advanced technologies, such as telemedicine, which enhance access to medical services for patients from remote regions or with limited mobility. Concepts for developing ethical and legal regulation of physician-patient relationships are linked to addressing global challenges, integrating the best global practices, and adapting them to national conditions. In conclusion, the creation of a comprehensive, patient-centered policy for the development of the healthcare system, based on ethics, law, and trust, is recommended to improve the quality of medical care and enhance the authority of the medical profession in society.

Regulatory and legal regulation of competition in the banking services market

The state of economic competition in the banking system of Ukraine is considered. A review of regulatory and legal documents of the European Union and the United States of America on the regulation of competitive relations in the economy and the prevention of monopolization in general, and the banking system in particular, is made. The opinions of researchers and institutions of leading countries of the world on the risks of violation of competition in the banking system are presented. The legal regulation of competitive relations in the legislation of Ukraine is analyzed. Provisions in the «general» competition legislation and «industry» banking legislation on the protection of economic competition are highlighted. The main types of violations of competition regulation and practices of unfair competition in the process of providing banking services are analyzed. The norms of legislation that should prevent such violations are studied. The complexity of the problem of protecting economic competition among banks is indicated, which includes not only direct violations of competition (unfair competition, misleading consumers of banking services, violations of advertising regulations), but also other unlawful administrative advantages for certain segments of the banking market (in particular, public sector banks). Separately, the issues of creating obstacles to business entities and achieving unlawful advantages in competition in the banking services market and regulatory regulation of banking secrecy and confidentiality of information in the financial system, which correlates with the provisions of antimonopoly legislation on the unlawful collection, disclosure and use of trade secrets, are highlighted. Regulatory acts and analytical documents of the Antimonopoly Committee of Ukraine on the current state of competition in the banking sector are considered. Conclusions are drawn regarding the lack of attention and influence of regulatory authorities on competition in the banking system, an increase in the level of concentration in the banking sector of Ukraine with a simultaneous decrease in the level of competition. The authors point out a direct violation, in their opinion, of the legislation of Ukraine on economic and competitive relations in certain regulatory acts of the executive branch.

PrivacyGuard: Exploring Hidden Cross-App Privacy Leakage Threats In IoT Apps

The increasing use of the Internet of Things (IoT) technology has made our lives convenient, however, it also poses new security and privacy threats. In this work, we study a new type of privacy threat enabled by cross-app chains built among multiple seemingly benign IoT apps. We find that interactions among apps could leak privacy-sensitive information, e.g., users' identification, location and tracking, activity patterns, etc. To tackle this challenge, we introduce PrivacyGuard, which extracts cross-app chains in the form of trigger-condition-action rules and identifies the corresponding privacy leakage risk with an inference probability. PrivacyGuard supports a fine-grained categorization of privacy threats to generate detailed alerts about privacy leakages. We evaluated PrivacyGuard on a dataset with 2,101 SmartApps, 2,788 IFTTT rules, and 2,086 OpenHAB rules, respectively. The results show that PrivacyGuard could uncover hidden privacy leaks that existing studies fail to detect. For example, 7.67% chains constructed by two seemingly benign IoT apps could leak at least one type of privacy information, while over 80% of the leaks involved privacy information regarding Localization & Tracking and Activity Profiling.

Navigating Privacy: A Global Comparative Analysis of Data Protection Laws

The increasing reliance on big data and artificial intelligence (AI) in the Fourth Industrial Revolution has raised significant concerns about individual privacy protection. This has led various countries to enact or amend privacy protection acts to address these concerns. However, there is a lack of comprehensive research comparing these laws across multiple countries, especially considering recent legislative developments. This study fills this gap by conducting a comparative analysis of privacy information protection acts in five major regions: the European Union (EU), the United States (focusing on California), China, Japan, and South Korea. The analysis explores the diverse approaches to privacy protection adopted by each region, influenced by their unique historical, political, and cultural contexts. For instance, the EU’s General Data Protection Regulation (GDPR) emphasizes individual rights influenced by historical abuses of personal information. At the same time, the California Consumer Privacy Act (CCPA) prioritizes consumer rights within a self‐regulatory framework, reflecting the state’s technology‐driven economy. The study also examines China’s Personal Information Protection Law (PIPL), which prioritizes national security; Japan’s Act on the Protection of Personal Information (APPI), which navigates the tension between individual privacy and societal norms; and South Korea’s Personal Information Protection Act (PIPA), which balances individual autonomy with a sense of community, reflecting Confucian values. By identifying specific limitations and areas for improvement in each region’s data protection laws, this study contributes to the ongoing discourse on international data privacy regulation. It offers valuable insights for policymakers and stakeholders seeking to navigate the complexities of the data economy while ensuring robust safeguards for individual privacy.

Anomaly Detection in IoT Network Traffic Using Bidirectional 3D Quasi‐Recurrent Neural Network Optimize With Coati Optimization Algorithm

ABSTRACTAnomaly detection in Internet of Things (IOT) network traffic involves identifying abnormal patterns or behaviors, enabling early detection of potential security threats or system malfunctions in the IOT ecosystem. IoT encompasses a variety of networks, addressing not only security challenges within sensor networks, the internet, and mobile communication networks but also specifically focusing on issues related to privacy protection, information management, network authentication, and access control. In this manuscript, Anomaly Detection in IoT Network Traffic using a Bidirectional 3D Quasi‐Recurrent Neural Network with Coati Optimization Algorithm (ADIOT‐B3DQRNN‐COA) is proposed. Initially, the input data are collected from the DS2OS Dataset. Then, the collected data is fed into pre‐processing utilizing an Implicit Unscented Particle Filter (IUPF). The IUPF is used to remove the invalid data. Subsequently, the preprocessed data are sent into the Archimedes optimization algorithm (AOA) to select features. Seven characteristics from the DS2OS dataset are chosen using AOA. The selected features are then fed into a Bidirectional 3D Quasi‐Recurrent Neural Network (Bi‐3DQRNN) to classify anomaly detection in an Internet of Things network into the following categories: data probing, malicious control, malicious operation, scan, spying, incorrect configuration, DOS attack, and normal. To guarantee accurate classification of anomaly detection in IoT networks, Bi‐3DQRNN generally does not express any adaptation of optimization algorithms for figuring out the best parameters. Hence, the Coati Optimization Algorithm (COA) to optimize Bi‐3DQRNN accurately classifies anomaly detection in the IOT network. The proposed ADIOT‐B3DQRNN‐COA approach is implemented in MATLAB. The performance of the proposed method was examined utilizing performance metrics like Accuracy, Computational Time, F‐measure, Precision, Recall, and ROC. The proposed ADIOT‐B3DQRNN‐COA approach contains 32.15%, 28.31%, and 24.18% higher accuracy, 28.69%, 33.54%, and 19.46% higher Precision, and 24.50%, 32.34%, and 24.18% lower Error Rate compared with existing methods, such as Anomaly Detection in IOT network depend on enhanced metaheuristic feature selection optimization and hybrid deep neural network (DNN‐ADIOT‐NTT), Investigation on the stability of DRNN depend on the oversampling method in IDIIOT (DRNN‐PSO‐IDIIOT) and light gradient boosting machine in Anomaly Detection in IOT network: concurrent feature selection with hyperparameter tuning concurrent feature selection with hyperparameter tuning (SMSG‐SCADA‐SNN) respectively.

Iterative local search for preserving data privacy

k-Anonymization is a popular approach for sharing datasets while preserving the privacy of personal and sensitive information. It ensures that each individual is indistinguishable from at least k-1 others in the anonymized dataset through data suppression or generalization, which inevitably leads to some information loss. The goal is to achieve k-anonymization with minimal information loss. This paper presents an efficient local search framework designed to address this challenge using arbitrary information loss metrics. The framework leverages anytime capabilities, allowing it to balance computation time and solution quality, thereby progressively improving the quality of the anonymized data. Our empirical evaluation shows that the proposed local search framework significantly reduces information loss compared to current state-of-the-art solutions, providing performance improvements of up to 54% and 43% w.r.t. the k-members and l-greedy heuristic solutions, the leading algorithms for large datasets. Additionally, our solution approach outperforms the Hun-garian-based solution, the best solution approach for small-size instances, by up to 4.7% on these instances.

Advantage distillation for quantum key distribution

Abstract Quantum key distribution promises information-theoretically secure communication, with data post-processing playing a vital role in extracting secure keys from raw data. While hardware advancements have significantly improved practical implementations, optimizing post-processing techniques offers a cost-effective avenue to enhance performance. Advantage distillation, which extends beyond standard information reconciliation and privacy amplification, has proven instrumental in various post-processing methods. However, the optimal post-processing remains an open question. Therefore, it is important to develop a comprehensive framework to encapsulate and enhance these existing methods. In this work, we propose an advantage distillation framework for quantum key distribution, generalizing and unifying existing key distillation protocols. Inspired by entanglement distillation, our framework not only integrates current techniques but also improves upon them. Notably, by employing classical linear codes, we achieve higher key rates, particularly in scenarios where one-time pad encryption is not used for post-processing. Our approach provides insights into existing protocols and offers a systematic way for further enhancements in quantum key distribution.

Examining user switching intention between generative AI platforms: A push-pull-mooring perspective

Due to the intense competition among generative AI platforms, users may switch from a platform to an alternative one. This may lead to user attrition and undermine the competitive advantage. The purpose of this research is to examine user switching intention between generative AI platforms based on the push-pull-mooring (PPM). Push factors include information hallucination, privacy risk, and dissatisfaction, while pull factors include perceived interactivity, perceived anthropomorphism, perceived personalization, and user experience. Mooring factor is switching costs. Both structural equation modeling (SEM) and fuzzy-set qualitative comparative analysis (fsQCA) were used to conduct data analysis. The results validate the hypotheses. The fsQCA found that dissatisfaction is a common core condition triggering switching intention.

Collection and sharing of health information in mental health and related systems in Australia: perspectives of people who access mental health services

BackgroundInformation sharing and information privacy are important issues in mental health services. Yet the perspectives of Australians who access mental health services about these issues are poorly understood. This article addresses the research question: What are the concerns of people who use mental health services about the collection and use of their health information in mental health and related systems in Australia?MethodsParticipants were 16 people who had accessed mental health services and were involved in a series of co-design workshops as part of a larger study. Focused group activities were conducted in which participants were asked to discuss and create a visual map to describe their collective experiences of sharing information relating to mental health and recovery. The data were analysed using qualitative content analysis and the coding techniques of constant comparative analysis.ResultsParticipants expressed a sense of unknowability about the information held and accessed by various organisations. They described providing information to a wide variety of entities, though not always in a free or informed way. They believed other information held about them had been generated and/or shared by people other than themselves, often without their knowledge or consent. Participants reported that both the collection and storage of this information involved risks for them, and they sometimes restricted the information they provided. To improve their comfort with the use of their information, participants recommended customised and transparent information collection; individuals’ access and input to their own information; and use of information only for their benefit.ConclusionsTrust is key to positive experiences of information provision and information sharing. Central to establishing trust are transparent practices that facilitate greater choice and control.

A Metric Differential Privacy Mechanism for Sentence Embeddings

Sentence embeddings represent the meaning of a given sentence using a fixed dimensional vector. Different approaches have been proposed in the Natural Language Processing (NLP) community for learning encoders that can produce accurate sentence embeddings that perform well for diverse downstream tasks that require sentence representations. Despite these prior work focusing mainly on creating accurate sentence embeddings, how to protect the privacy of the sensitive information contained in the sentences remains an unexplored research problem. In this paper, we propose, Covering Metric Analytic Gaussian (CMAG), a covering metric Differential Privacy (DP) mechanism for sentence embeddings such that minimal random noise is added to a set of sentence embeddings produced by an encoder to protect the private information expressed in those sentences. Given a sentence embedding s, CMAG considers the Mahalanobis distance between s and the other sentence embeddings s′ in the local neighbourhood of s to determine the minimal amount of random noise that must be added to s to obtain provable metric DP guarantees. Experimental results show that the proposed DP mechanism protects private information better than previously proposed DP mechanisms, while reporting good performance in a broad range of downstream NLP tasks.

Climate change and state interference: the case of privacy

AbstractClimate change is one of the most important issues we are currently facing. There are many ways in which states can fight climate change. Some of them involve interfering with citizens’ personal lives. The question of whether such interference is justified is under-explored in philosophy. This paper focuses on a specific aspect of people’s personal lives, namely their informational privacy. It discusses the question of whether, given certain empirical assumptions, it is proportional of the state to risk its citizens’ privacy or to risk infringing its citizens’ right to privacy to fight climate change. The main claim this paper argues for is that if fighting climate change and protecting our privacy conflict, we have good reason to fight climate change rather than protect our privacy.