Abstract
Graph structured data can be found in many domains and applications. Analysis of such data can give valuable insights. Frequent subgraph discovery, the problem of finding the set of subgraphs that is frequent among the underlying database of graphs, has attracted a lot of recent attention. Many algorithms have been proposed to solve this problem. However, all assume that the entire set of graphs is centralized at a single site, which is not true in a lot of cases. Furthermore, in a lot of interesting applications, the data is sensitive (for example, drug discovery, clique detection, etc). In this paper, we address the problem of privacy-preserving subgraph discovery. We propose a flexible approach that can utilize any underlying frequent subgraph discovery algorithm and uses cryptographic primitives to preserve privacy. The comprehensive experimental evaluation validates the feasibility of our approach.