Abstract:There exists a growing concern about the privacy threats inherently encoded in the increasingly available behavioral data. In this paper, we focus on online service providers over which users reveal preferences (e.g., ratings or reviews) as publicly available behavioral data. We model users' strategic interactions in making these decisions as quadratic games on networks, where a user's payoff depends not only on their actions but also on their neighbors in a social interaction network. Based on the assumption that the observed preferences correspond to the Nash equilibrium of the game, we investigate whether eavesdroppers or competitors having access to such behavioral data could potentially infer or even thoroughly learn the hidden social interaction network from the observed user actions, leading to privacy risks at a system level for the platform. We introduce three notions of privacy risks on networks: fundamental privacy, conditional privacy, and practical privacy. We establish necessary and sufficient conditions for fundamental and conditional privacy of the social interaction structure, suggesting that at the system level, such interaction structure is not facing a critical risk of being fully exposed regardless of the amount of data available. To study practical privacy, we further propose and analyze a novel learning framework by solving a joint optimization problem for the network structure and the individual marginal benefits in the game. Extensive experiments on synthetic and real-world data demonstrate the effectiveness of the proposed framework, hence illustrating that behavioral data indeed present substantial privacy risks in practice. Broadly, this study enriches the network privacy literature and provides implications for online platforms by revealing the unexpected consequence of leaking network connections due to public consumer revealed preferences.

Diversity, Topology, and the Risk of Node Re-identification in Labeled Social Graphs

Devil in Disguise: Breaching Graph Neural Networks Privacy through Infiltration

Privacy Risks of Social Interaction Structure: Network Learning in Quadratic Games

Privacy Risk in Anonymized Heterogeneous Information Networks

Does Black-box Attribute Inference Attacks on Graph Neural Networks Constitute Privacy Risk?

Privacy Protection in Social Networks Using L-Diversity

Local Differential Privacy in Graph Neural Networks: a Reconstruction Approach

Applying l-Diversity in anonymizing collaborative social network

Attribute-Enhanced De-anonymization of Online Social Networks

De-Anonymizing Dynamic Social Networks.

Unveiling Privacy Vulnerabilities: Investigating the Role of Structure in Graph Data

The effect of distant connections on node anonymity in complex networks

On Learning Cluster Coefficient of Private Networks

De-SAG: on the De-Anonymization of Structure-Attribute Graph Data

Structure-Attribute-Based Social Network Deanonymization with Spectral Graph Partitioning

On Anonymization of Multi-graphs.

Privacy-Utility Trade-offs in Neural Networks for Medical Population Graphs: Insights from Differential Privacy and Graph Structure

A Privacy Preservation Method for Attributed Social Network Based on Negative Representation of Information

De-anonymization of Social Networks: the Power of Collectiveness

Blind De-anonymization Attacks using Social Networks