A privacy-preserving decentralized credit scoring method based on multi-party information

2023 | Decision Support Systems | Citations: 0

Authors: He, Haoran; Wang, Zhao; Jain, Hemant; Jiang, Cuiqing; Yang, Shanlin

Abstract: With society's wide-scale adoption of information technology, significant inform ... Expand

Abstract: With society's wide-scale adoption of information technology, significant information about borrowers is distributed across various parties, information that can be jointly used to improve credit scoring. However, use of such information faces many challenges, such as the problems of preserving privacy and information redun­ dancy. To address these challenges in leveraging multi-source information for credit scoring, we propose a decentralized multi-party method based on logistic regression. Specifically, we formulate a logistic regression model using the vertical federated learning paradigm. To preserve data privacy during multi-party collaborative model training, we use additively homomorphic encryption based on the second-order Taylor series expansion of the loss function and its gradient. To address information redundancy and to improve the performance of the credit scoring model, we use the proposed HE-DPGD algorithm to estimate the distributed coefficients in a privacy-preserving setting. Empirical evaluation shows that the proposed method can leverage information from multiple parties securely and effectively. Collapse

Topics: privacy homomorphic encryption logistics management online lending missing data

Methods: computational algorithm machine learning federated learning logistic regression experiment

Reidentification Risk in Panel Data: Protecting for k-Anonymity

2023 | Information Systems Research | Citations: 0

Authors: Li, Shaobo; Schneider, Matthew J.; Yu, Yan; Gupta, Sachin

Abstract: We consider the risk of reidentification of panelists in marketing research data ... Expand

Abstract: We consider the risk of reidentification of panelists in marketing research data that are widely used to obtain insights into buyer behavior and to develop marketing strategy. We find that 17%–94% of the panelists in 15 frequently bought consumer goods categories are subject to high risk of reidentification through a potential record linkage attack based on their unique purchasing histories even when their identities are anonymized. We first demonstrate that the risk of reidentification is vastly understated by unicity, the conventional measure. Instead, we propose a new measure of reidentification risk, termed sno-unicity, which accounts for the longitudinal nature of panel data, and show that it is much larger than unicity. To protect the privacy of panelists, we consider the well-known privacy notion of k-anonymity and develop a new approach called graph-based minimum movement k-anonymization (k-MM) that is designed especially for panel data. The proposed k-MM approach can be formulated as an optimization problem in which the objective is to minimally distort variables in the original data based on weights that users prespecify corresponding to their use case. We further show how our approach can be extended to achieve l-diversity. We apply the k-MM approach to two different panel data sets that are widely used in marketing research. To achieve a given privacy level, compared with several benchmark protection methods, the protected data from our method result in the least distortion in inferences about key marketing metrics, such as brand market shares, share of category requirements, brand switching rates, and marketing-mix parameters estimated from a hierarchical Bayesian brand choice model. History: Param Singh, Senior Editor and Associate Editor. Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2022.1169. Collapse

Topics: privacy marketing management differential privacy data protection record linkage

Methods: longitudinal research optimization model descriptive statistic simulation cluster analysis

Understanding the Digital Resilience of Physicians during the COVID-19 Pandemic: An Empirical Study

2023 | Management Information Systems Quarterly | Citations: 0

Authors: Liu, Yinghao; Xu, Xin; Jin, Yong; Deng, Honglin

Abstract: The COVID-19 pandemic has underscored the urgent need for healthcare entities t ... Expand

Abstract: The COVID-19 pandemic has underscored the urgent need for healthcare entities to develop resilient strategies to cope with disruptions caused by the pandemic. This study focuses on the digital resilience of certified physicians who adopted an online healthcare community (OHC) to acquire patients and conduct telemedicine services during the pandemic. We synthesize the resilience literature and identify two effects of digital resilience-the resistance effect and the recovery effect. We use a proprietary dataset that matches online and offline data sources to study the digital resilience of physicians. A difference-in-differences (DID) analysis shows that physicians who adopted an OHC had strong resistance and recovery effects during the pandemic. Remarkably, after the COVID-19 outbreak, these physicians had 35.0% less reduction in medical consultations in the immediate period and 31.0% more bounce-back in the subsequent period as compared to physicians who did not adopt the OHC. We further analyze the sources of physicians' digital resilience by distinguishing between new and existing patients from both online and offline channels. Our subgroup analysis shows that, in general, digital resilience is more pronounced when physicians have a higher online reputation rating or have more positive interactions with patients on the OHC platform, providing further support for the mechanisms underlying digital resilience. Our research has significant theoretical and managerial implications beyond the context of the pandemic. Collapse

Topics: pandemic electronic health business continuity planning disaster recovery information system use

Methods: longitudinal research experimental group longitudinal survey augmented Dickey–Fuller test time series analysis

Open Source Collaboration in Digital Entrepreneurship

2022 | Organization Science | Citations: 3

Authors: Lin, Yu-Kai; Maruping, Likoebe M.

Abstract: Emerging digital technologies give rise to digital entrepreneurship and the wide ... Expand

Abstract: Emerging digital technologies give rise to digital entrepreneurship and the widespread phenomenon of open source collaboration (OSC) on GitHub for entrepreneurial pursuits. Although openness is a common theme in digital entrepreneurship, it is unclear how digital startups—that is, startups that that have digital artifacts at the core of their business model for value creation and capture—actually realize value from their OSC engagement. We develop a theoretical framework to explain how the engagement in OSC may affect the value of digital startups and how the effect is contingent on the stage of venture maturity (conception, commercialization, or growth) and the mode of OSC engagement (inbound or outbound). In analyses that pool 17,552 matched digital startups with monthly panel observations between 2008 and 2017, we find digital startups in the conception and commercialization stages benefit more from inbound OSC whereas the ones in the growth stage benefit more from outbound OSC. As digital startups increasingly use OSC for ideation, experimentation, and scaling, our contribution is to show whether, when, and how knowledge flows through OSC might affect the value of digital startups. We discuss implications for research on organizing for digital entrepreneurship as well as open innovation. Collapse

Topics: digital entrepreneurship open innovation innovation management value creation open source

Methods: theory development longitudinal research experimental group descriptive statistic regression analysis method

Theories: organizational theory

Data quality in recommender systems: the impact of completeness of item content data on prediction accuracy of recommender systems

2021 | Electronic Markets | Citations: 1

Authors: Heinrich, Bernd; Hopf, Marcus; Lohninger, Daniel; Schiller, Alexander; Szubartowicz, Michael

Abstract: Recommender systems strive to guide users, especially in the field of e-commerce ... Expand

Abstract: Recommender systems strive to guide users, especially in the field of e-commerce, to their individually best choice when a large number of alternatives is available. In general, literature suggests that the quality of data which a recommender system is based on may have important impact on recommendation quality. In this paper, we focus on the data quality dimension completeness of item content data (i.e., features of items and their feature values) and investigate its impact on the prediction accuracy of recommender systems. In particular, we examine the increase in completeness per item, per user and per feature as moderators for this impact. To this end, we present a theoretical model based on the literature and derive ten hypotheses. We test these hypotheses on two real-world data sets, one from two leading web portals for restaurant reviews and another one from a movie review portal. The results strongly support that, in general, the prediction accuracy is positively influenced by increased completeness. However, the results also reveal, contrary to existing literature, that among others increasing completeness by adding features which differ significantly from already existing features (i.e., a high diversity) does not positively influence the prediction accuracy of recommender systems. Collapse

Topics: recommender system data quality missing data information portal decision support system

Methods: statistical hypothesis test regression analysis method data imputation Student's t-test parametric test

Anomaly-Based Duplicate Detection: A Probabilistic Approach

2019 | DESRIST | Citations: 0

Authors: Obermeier, Andreas

Abstract: The importance of identifying records in databases that refer to the same real-w ... Expand

Abstract: The importance of identifying records in databases that refer to the same real-world entity (“duplicate detection”) has been recognized in both research and practice. However, existing supervised approaches for duplicate detection need training data with labeled instances of duplicates and non-duplicates, which is often costly and time-consuming to generate. On the contrary, unsupervised approaches can forego such training data but may suffer from limiting assumptions (e.g., monotonicity) and providing less reliable results. To address the issue of generating high-quality results using easy to acquire duplicate-free training data only, we propose a probabilistic approach for anomaly-based duplicate detection. Duplicates exhibit specific characteristics which differ significantly from the characteristics of non-duplicates and therefore represent anomalies. Based on the grade of anomaly compared to duplicate-free training data, our approach assigns the probability of being a duplicate to each analyzed pair of records while avoiding limiting assumptions (of existing approaches). We demonstrate the practical applicability and effectiveness of our approach in a real-world setting by analyzing customer master data of a German insurer. The evaluation shows that the results provided by the approach are reliable and useful for decision support and can outperform even fully supervised state-of-the-art approaches for duplicate detection. Collapse

Topics: Python record linkage knowledge base database system decision making

Methods: design artifact support vector machine machine learning literature study design science

Theories: probability theory decision theory

The Long-term Diffusion of Digital Platforms — An Agent-based Model

2019 | International Conference on Information Systems | Citations: 0

Authors: Schalowski, Jan; Barrot, Christian

Abstract: In recent years, many industries have experienced the rise of digital platforms ... Expand

Abstract: In recent years, many industries have experienced the rise of digital platforms (e.g., eBay, Uber, or Takeaway.com). A common characteristic of these concepts is that they focus on fragmented markets populated by many small firms, which often show a high fluctuation. However, established diffusion models based on Bass (1969) do not account for fluctuation in the market potential, although the exit of adopters and the entry of new firms could change the diffusion curve significantly. Thus, we propose an extension of the Bass Model to account for the exit and entry of (potential) adopters and empirically test this framework in a real-world setting. Using two decades of adopter data of leading digital platforms and information on the complete market potential, we employ agent-based models to analyze the effects of fluctuation on the platform diffusion. Initial results confirm the existence of high fluctuation and indicate relevant impacts on the diffusion curve. Collapse

Topics: digital platform social influence innovation management Python viral marketing

Methods: agent based simulation simulation

Digression and Value Concatenation to Enable Privacy-Preserving Regression

2014 | Management Information Systems Quarterly | Citations: 0

Authors: Xiao-Bai Li; Sarkar, Sumit

Abstract: Regression techniques can be used not only for legitimate data analysis, but als ... Expand

Abstract: Regression techniques can be used not only for legitimate data analysis, but also to infer private information about individuals. In this paper, we demonstrate that regression trees, a popular data-analysis and data-mining technique, can be used to effectively reveal individuals’ sensitive data. This problem, which we call a regression attack, has not been addressed in the data privacy literature, and existing privacy-preserving techniques are not appropriate in coping with this problem. We propose a new approach to counter regression attacks. To protect against privacy disclosure, our approach introduces a novel measure, called digression, which assesses the sensitive value disclosure risk in the process of building a regression tree model. Specifically, we develop an algorithm that uses the measure for pruning the tree to limit disclosure of sensitive data. We also propose a dynamic value-concatenation method for anonymizing data, which better preserves data utility than a user-defined generalization scheme commonly used in existing approaches. Our approach can be used for anonymizing both numeric and categorical data. An experimental study is conducted using real-world financial, economic, and healthcare data. The results of the experiments demonstrate that the proposed approach is very effective in protecting data privacy while preserving data quality for research and analysis. Collapse

Topics: privacy differential privacy data quality healthcare data record linkage

Methods: computational algorithm experiment regression analysis method linear regression analysis archival research

Protecting Privacy Against Record Linkage Disclosure: A Bounded Swapping Approach for Numeric Data

2011 | Information Systems Research | Citations: 0

Authors: Xiao-Bai Li; Sarkar, Sumit

Abstract: Record linkage techniques have been widely used in areas such as antiterrorism, ... Expand

Abstract: Record linkage techniques have been widely used in areas such as antiterrorism, crime analysis, epidemiologic research, and database marketing. On the other hand, such techniques are also being increasingly used for identity matching that leads to the disclosure of private information. These techniques can be used to effectively reidentify records even in deidentified data. Consequently, the use of such techniques can lead to individual privacy being severely eroded. Our study addresses this important issue and provides a solution to resolve the conflict between privacy protection and data utility. We propose a data-masking method for protecting private information against record linkage disclosure that preserves the statistical properties of the data for legitimate analysis. Our method recursively partitions a data set into smaller subsets such that data records within each subset are more homogeneous after each partition. The partition is made orthogonal to the maximum variance dimension represented by the first principal component in each partitioned set. The attribute values of a record in a subset are then masked using a double-bounded swapping method. The proposed method, which we call multivariate swapping trees, is nonparametric in nature and does not require any assumptions about statistical distributions of the original data. Experiments conducted on real-world data sets demonstrate that the proposed approach significantly outperforms existing methods in terms of both preventing identity disclosure and preserving data quality. Collapse

Topics: record linkage privacy data quality website

Methods: computational algorithm experiment artificial neural network simulation linear regression analysis

DATA CLUSTERING AND MICRO-PERTURBATION FOR PRIVACY-PRESERVING DATA SHARING AND ANALYSIS

2010 | International Conference on Information Systems | Citations: 0

Authors: Li, Xiao-Bai; Sarkar, Sumit

Abstract: Clustering-based data masking approaches are widely used for privacy-preserving ... Expand

Abstract: Clustering-based data masking approaches are widely used for privacy-preserving data sharing and data mining. Existing approaches, however, cannot cope with the situation where confidential attributes are categorical. For numeric data, these approaches are also unable to preserve important statistical properties such as variance and covariance of the data. We propose a new approach that handles these problems effectively. The proposed approach adopts a minimum spanning tree technique for clustering data and a micro-perturbation method for masking data. Our approach is novel in that it (i) incorporates an entropy-based measure, which represents the disclosure risk of the categorical confidential attribute, into the traditional distance measure used for clustering in an innovative way; and (ii) introduces the notion of cluster-level microperturbation (as opposed to conventional micro-aggregation) for masking data, to preserve the statistical properties of the data. We provide both analytical and empirical justification for the proposed methodology. Collapse

Topics: privacy record linkage healthcare data differential privacy

Methods: cluster analysis computational algorithm experiment ANOVA parametric test