Towards Blockchain-Assisted Privacy-Aware Data Sharing For Edge Intelligence: A Smart Healthcare Perspective

Youyang Qu, Lichuan Ma, Wenjie Ye, Xuemeng Zhai, Shui Yu, Yunfeng Li, David B. SmitharXiv 2023 · DOI · OpenReview

TL;DR

This paper introduces a novel Community-based Differential Privacy (C-DP) model for smart healthcare networks (SHNs). It personalizes privacy levels based on user trust within communities, uses blockchain to prevent data falsification, and employs a noise-decoupling mechanism to thwart linkage attacks, offering a more robust and adaptable privacy solution than existing methods.

What Problem Does This Paper Address?

The proliferation of smart healthcare devices generates vast amounts of sensitive health data. Sharing this data is crucial for accurate diagnoses and personalized treatments, but it raises significant privacy concerns. Existing methods struggle with:

  • Linkage Attacks: Adversaries can combine data from multiple sources to re-identify individuals and infer sensitive information.
  • Poisoning Attacks: Malicious actors can inject or alter data, leading to misdiagnoses and potential harm.
  • One-Size-Fits-All Privacy: Traditional differential privacy offers a constant level of protection for all users, which is inefficient and may not adequately protect highly sensitive data or provide enough utility for less sensitive data.
  • Lack of Trust-Based Personalization: Existing personalization methods (like virtual online distance) are often insufficient or difficult to implement.

This paper aims to address these challenges by creating a privacy-preserving data sharing framework for SHNs that is personalized, secure, and resilient to common attacks.

Methods

The proposed solution, Community-based Differential Privacy (C-DP), integrates several key techniques:

  • Community Detection: Users are grouped into communities based on their relationships and interactions. The paper uses an edge-pair based community detection algorithm that models SHNs as graphs. This method aims to avoid the "overlapped nodes problem" where users might belong to multiple communities, which could lead to inconsistent privacy levels.
  • Trust-Based Personalization: User trust within a community is measured by community density (how interconnected users are). This density is then mapped to a personalized privacy protection level (represented by epsilon, $\epsilon$) using a modified sigmoid function. Higher density implies higher trust, leading to a lower privacy protection level (more data utility) and vice versa.
  • Noise Decoupling: To prevent linkage attacks, a noise correlation decoupling mechanism is introduced. This mechanism, based on a Markov stochastic process, ensures that the injected noise is uncorrelated across different data points, making it harder for attackers to infer sensitive information by combining noisy data.
  • Blockchain Integration: A consortium blockchain is employed to store and manage the differentially private health data. This provides immutability and integrity, making it extremely difficult for adversaries to falsify data. The blockchain also facilitates secure data access control.
  • Differential Privacy Mechanisms: Both Laplace (for numerical data) and Exponential (for textual data) mechanisms are used to inject privacy-preserving noise, adhering to $\epsilon$-differential privacy guarantees.

Datasets

The authors evaluated their proposed system using two real-world datasets:

  • Doximity Dataset: Obtained from Doximity's Developer API, this dataset consists of approximately 2,122 nodes and 14,389 edges.
  • HealthTap Dataset: This dataset was built from the HealthTap API and includes approximately 1,325 nodes and 5,231 edges.

These datasets represent social networks for doctors, allowing the researchers to model community structures and relationships relevant to healthcare.

Key Insights and Ideas

  • Community Density as a Trust Metric: The paper effectively demonstrates that community density is a practical and intuitive proxy for user trust in SHNs, enabling personalized privacy.
  • Blockchain for Data Integrity: Integrating a consortium blockchain is a robust way to ensure that once data is shared, it cannot be tampered with, directly addressing the threat of data falsification.
  • Noise Decoupling is Crucial for Linkage Attack Mitigation: The proposed mechanism effectively breaks the correlation between noisy data points, which is a key vulnerability exploited by linkage attacks.
  • Personalization Outperforms Fixed Privacy: The experimental results suggest that personalized privacy levels, tailored to community trust, offer a better balance between privacy and utility compared to uniform privacy protection.
  • Edge-Pair Community Detection: Using edges to define communities helps to avoid the ambiguity of overlapping nodes, leading to more consistent privacy assignments.

Key Results

  • Attack Mitigation: The proposed C-DP model effectively mitigates linkage attacks and poisoning attacks. Experiments showed that poisoning attacks were "totally mitigated" when the number of users in a community exceeded 30.
  • Privacy-Utility Trade-off: The system achieves a good balance between privacy protection and data utility. As the privacy level ($\epsilon$) increases, data utility generally increases, and vice versa, demonstrating a well-managed trade-off.
  • Performance Comparison: The C-DP model, especially when combined with blockchain, showed superior performance in terms of security and privacy protection compared to traditional DP and personalized DP (P-DP) methods.
  • Scalability: The computation overhead for optimization was shown to converge to a specific value as the number of communities increases, suggesting good scalability.

Limitations and Future Work

  • Community Detection Granularity: The authors acknowledge that controlling the granularity of community detection is an important procedure but not easy.
  • Theoretical Foundation: Further clarification of the theoretical foundation for the optimized trade-off between privacy and utility is needed.
  • Game Theory for Adversary Modeling: Future work will involve using game theory to better model the confrontation between data holders and adversaries, aiming for a more balanced privacy-utility trade-off.
  • Federated Learning Integration: The integration of federated learning is being explored to further enhance privacy protection in SHNs.

Summary updated

Abstract

The popularization of intelligent healthcare devices and big data analytics significantly boosts the development of smart healthcare networks (SHNs). To enhance the precision of diagnosis, different participants in SHNs share health data that contains sensitive information. Therefore, the data exchange process raises privacy concerns, especially when the integration of health data from multiple sources (linkage attack) results in further leakage. Linkage attack is a type of dominant attack in the privacy domain, which can leverage various data sources for private data mining. Furthermore, adversaries launch poisoning attacks to falsify the health data, which leads to misdiagnosing or even physical damage. To protect private health data, we propose a personalized differential privacy model based on the trust levels among users. The trust is evaluated by a defined community density, while the corresponding privacy protection level is mapped to controllable randomized noise constrained by differential privacy. To avoid linkage attacks in personalized differential privacy, we designed a noise correlation decoupling mechanism using a Markov stochastic process. In addition, we build the community model on a blockchain, which can mitigate the risk of poisoning attacks during differentially private data transmission over SHNs. To testify the effectiveness and superiority of the proposed approach, we conduct extensive experiments on benchmark datasets.
BibTeX
@misc{qu2023towards,
  author = {Youyang Qu and Lichuan Ma and Wenjie Ye and Xuemeng Zhai and Shui Yu and Yunfeng Li and David B. Smith},
  title = {{Towards Blockchain-Assisted Privacy-Aware Data Sharing For Edge Intelligence: A Smart Healthcare Perspective}},
  year = {2023},
  eprint = {2306.16630},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2306.16630},
}
Figure