The article discusses the issue of privacy in graph embedding, a powerful tool for learning latent representations for nodes in a graph. The authors highlight the need for privacy-preserving graph embedding algorithms due to the sensitive information involved in most real-world graphs. They propose a new framework, LDP-GE, that satisfies local differential privacy (LDP), a paradigm that ensures stronger privacy guarantees for users in data aggregation scenarios. The LDP-GE framework includes an LDP mechanism to obfuscate node data and uses personalized PageRank as the proximity measure to learn node representations. The research concludes that LDP-GE achieves favorable privacy-utility trade-offs and outperforms existing approaches in both node classification and link prediction tasks.
Publication date: 19 Oct 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2310.11060
