Vertical federated learning (VFL), a variant of federated learning, has recently attracted increasing attention. An active party having the true labels jointly trains a model with other parties (referred to as passive parties) in order to use more features to achieve higher model accuracy. During the prediction phase, all the parties collaboratively compute the predicted confidence scores of each target record and the results will be finally returned to the active party. However, a recent study by Luo et al. [28] pointed out that the active party can use these confidence scores to reconstruct passive-party features and cause severe privacy leakage. In this paper, we conduct a comprehensive analysis of privacy leakage in VFL frameworks during the prediction phase. Our study improves on previous work [28] regarding two aspects. We first design a general gradient-based reconstruction attack framework that can be flexibly applied to simple logistic regression models as well as multi-layer neural networks. Moreover, besides performing the attack under the white-box setting, we give the first attempt to conduct the attack under the black-box setting. Extensive experiments on a number of real-world datasets show that our proposed attack is effective under different settings and can achieve at best twice or thrice of a reduction of attack error compared to previous work [28]. We further analyze a list of potential mitigation approaches and compare their privacy-utility performances. Experimental results demonstrate that privacy leakage from the confidence scores is a substantial privacy risk in VFL frameworks during the prediction phase, which cannot be simply solved by crypto-based confidentiality approaches. On the other hand, processing the confidence scores with information compression and randomization approaches can provide strengthened privacy protection.

Volunteer moderators have the power to shape society through their influence on online discourse. However, the growing scale of online interactions increasingly presents significant hurdles for meaningful moderation. Furthermore, there are only limited tools available to assist volunteers with their work. Our work aims to meaningfully explore the potential of AI-driven, automated moderation tools for social media to assist volunteer moderators. One key aspect is to investigate the degree to which tools must become personalizable and context-sensitive in order to not just delete unsavory content and ban trolls, but to adapt to the millions of online communities on social media mega-platforms that rely on volunteer moderation. In this study, we conduct semi-structured interviews with 26 Facebook Group moderators in order to better understand moderation tasks and their associated challenges. Through qualitative analysis of the interview data, we identify and address the most pressing themes in the challenges they face daily. Using interview insights, we conceptualize three tools with automated features that assist them in their most challenging tasks and problems. We then evaluate the tools for usability and acceptance using a survey drawing on the technology acceptance literature with 22 of the same moderators. Qualitative and descriptive analyses of the survey data show that context-sensitive, agency-maintaining tools in addition to trial experience are key to mass adoption by volunteer moderators in order to build trust in the validity of the moderation technology.

The Chinese social credit system (SCS) is a digital sociotechnical credit system that rewards and sanctions the economic and social behaviors of individuals and companies. This article uses classic social-control theories—the shaming theory and the labeling theory—to analyze the SCS, thereby contributing to a better understanding of the Chinese social-control approach to the digital transformation of society. Our research relies not only on government documents and media reports, but also on first-hand data collected from in-depth interviews conducted in China. We found that the perceived effectiveness of the shaming and labeling mechanisms is enhanced by the design of the SCS framework and the assistance of digital technology but weakened by a lack of transparency and questionable justification criteria, as well as privacy and fairness concerns.

For almost three years, the General Data Protection Regulation (GDPR) has been granting citizens of the European Union the right to obtain personal data from companies and to transfer these data to another company. The so-called Right to Data Portability (RtDP) promises to significantly reduce switching costs for consumers in digital service markets, provided that its potential is effectively translated into reality. Thus, of all the consumer rights in the GDPR, the RtDP has the potential to be the one with the most significant implications for digital markets and privacy. However, our research shows that the RtDP is barely known among consumers and can currently only be implemented in a fragmented manner—especially with regard to the direct transfer of data between online service providers. We discuss several ways to improve the implementation of this right in the present article.

Integer overflows have threatened software applications for decades. Thus, in this paper, we propose a novel technique to provide automatic repairs of integer overflows in C source code. Our technique, based on static symbolic execution, fuses detection, repair generation and validation. This technique is implemented in a prototype named IntRepair. We applied IntRepair to 2,052 C programs (approx. 1 million lines of code) contained in SAMATE's Juliet test suite and 50 synthesized programs that range up to 20 KLOC. Our experimental results show that IntRepair is able to effectively detect integer overflows and successfully repair them, while only increasing the source code (LOC) and binary (Kb) size by around 1%, respectively. Furthermore, we present the results of a user study with 30 participants showing that IntRepair repairs are more efficient than manual repairs.