Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

P4GCN: Vertical Federated Social Recommendation with Privacy-Preserving Two-Party Graph Convolution Network

Zheng Wang, Wanwan Wang, Yimin Huang, Zhaopeng Peng, Ziqi Yang, Ming Yao, Cheng Wang, Xiaoliang Fan

TL;DR

P4GCN addresses the challenge of leveraging social information for recommendations when social data are inaccessible due to privacy and business constraints. It introduces a vertical federated GNN framework with a Sandwich Encryption module and differential privacy via a matrix Gaussian mechanism to securely perform social aggregation without exposing sensitive inputs, while employing a fusion layer to retain utility. The method achieves competitive or superior RMSE/MAE across four real-world datasets and demonstrates favorable privacy-utility trade-offs, supported by theoretical privacy guarantees under honest-but-curious assumptions. Practically, P4GCN enables privacy-preserving cross-party social recommendations with reasonable communication costs and shows potential to augment existing local recommender systems through a plug-in P4 layer.

Abstract

In recent years, graph neural networks (GNNs) have been commonly utilized for social recommendation systems. However, real-world scenarios often present challenges related to user privacy and business constraints, inhibiting direct access to valuable social information from other platforms. While many existing methods have tackled matrix factorization-based social recommendations without direct social data access, developing GNN-based federated social recommendation models under similar conditions remains largely unexplored. To address this issue, we propose a novel vertical federated social recommendation method leveraging privacy-preserving two-party graph convolution networks (P4GCN) to enhance recommendation accuracy without requiring direct access to sensitive social information. First, we introduce a Sandwich-Encryption module to ensure comprehensive data privacy during the collaborative computing process. Second, we provide a thorough theoretical analysis of the privacy guarantees, considering the participation of both curious and honest parties. Extensive experiments on four real-world datasets demonstrate that P4GCN outperforms state-of-the-art methods in terms of recommendation accuracy.

P4GCN: Vertical Federated Social Recommendation with Privacy-Preserving Two-Party Graph Convolution Network

TL;DR

P4GCN addresses the challenge of leveraging social information for recommendations when social data are inaccessible due to privacy and business constraints. It introduces a vertical federated GNN framework with a Sandwich Encryption module and differential privacy via a matrix Gaussian mechanism to securely perform social aggregation without exposing sensitive inputs, while employing a fusion layer to retain utility. The method achieves competitive or superior RMSE/MAE across four real-world datasets and demonstrates favorable privacy-utility trade-offs, supported by theoretical privacy guarantees under honest-but-curious assumptions. Practically, P4GCN enables privacy-preserving cross-party social recommendations with reasonable communication costs and shows potential to augment existing local recommender systems through a plug-in P4 layer.

Abstract

In recent years, graph neural networks (GNNs) have been commonly utilized for social recommendation systems. However, real-world scenarios often present challenges related to user privacy and business constraints, inhibiting direct access to valuable social information from other platforms. While many existing methods have tackled matrix factorization-based social recommendations without direct social data access, developing GNN-based federated social recommendation models under similar conditions remains largely unexplored. To address this issue, we propose a novel vertical federated social recommendation method leveraging privacy-preserving two-party graph convolution networks (P4GCN) to enhance recommendation accuracy without requiring direct access to sensitive social information. First, we introduce a Sandwich-Encryption module to ensure comprehensive data privacy during the collaborative computing process. Second, we provide a thorough theoretical analysis of the privacy guarantees, considering the participation of both curious and honest parties. Extensive experiments on four real-world datasets demonstrate that P4GCN outperforms state-of-the-art methods in terms of recommendation accuracy.

Paper Structure

This paper contains 54 sections, 4 theorems, 10 equations, 6 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related works
  3. Social recommendation
  4. Federated recommendation
  5. Problem formulation
  6. Methodology
  7. Motivation
  8. Sandwich encryption
  9. Solution to R1
  10. Solution to R2
  11. Database-level protection
  12. Element-level protection
  13. P4GCN
  14. Architecture
  15. Overview.
  16. ...and 39 more sections

Key Result

theorem 1

Given $\bold J=\bold L\bold M\bold N$ where all matrices are not zero matrices, there exists infinite combinations of $\bold N'\ne \bold N, \bold L'\ne\bold L$ such that $\bold J=\bold L'\bold M\bold N'$.

Figures (6)

  • Figure 1: The example of vertical federated social recommendation with inaccessible social data.
  • Figure 2: The framework of the Sandwich Encryption that computes arbitrary triple-matrix multiplication $\textbf{J=LMN}$ in a privacy-preserved way.
  • Figure 3: The training workflow of the proposed P4GCN
  • Figure 4: The model performance RMSE and MAE of P4GCN w/w.o. fusion layer v.s. privacy budget $\epsilon$.
  • Figure 5: The impact of social aggregation degree $\beta$ v.s. MAE
  • ...and 1 more figures

Theorems & Definitions (6)

  • theorem 1
  • definition 1: analytic Matrix Gaussian Mechanism du2023dp
  • definition 2: Matrix Gaussian Distributionyang2021imgm
  • lemma 1: DP of aMGM du2023dp
  • lemma 2
  • theorem 2