Graph-Based Ranking of Quasi-Identifier Combinations for Privacy Risk Assessment: A PageRank-Driven Framework

Abstract

Listen

Translate article

In the era of data-driven applications, ensuring privacy through effective identification of quasi- identifiers has become a critical challenge. This research focuses on the novel task of rank- ing combinations of quasi-identifiers within a knowledge graph to enhance privacy-preserving mechanisms. Leveraging PageRank as the foundational centrality measure, we introduce modi- fications such as logarithmic scaling to mitigate bias towards general entity nodes that are high in centrality and combination-based averaging to capture multi-entity interactions. The proposed approach not only identifies sensitive quasiidentifier combinations but also addresses limitations of traditional centrality measures such as degree centrality by balancing global significance with local contextual importance. To demonstrate its efficacy, we construct a domain-specific knowl- edge graph from the IMDb dataset, representing entities such as movies, actors, and genres, and validate our method through comparative analyses with existing centrality metrics. Our results highlight the improved accuracy and contextual relevance of the rankings, showcasing the poten- tial for applications in privacy preservation, data anonymization, and knowledge-based analytics. This work bridges the gap between graph-based centrality measures and domain-specific privacy requirements, offering a robust framework for sensitive information management.