Secure Similar Document Detection: Optimized Computation using the Jaccard Coefficient
Document Type
Conference Proceeding
Publication Date
11-28-2018
Abstract
Secure Similar Document Detection (SSDD) problem considers two parties, each holding a private document, who want to compute the similarity between their documents without leaking the document contents to one another. This is a unique problem whose applications span across a variety of domains, including the medical field, military intelligence, and academia. In this paper, we aim to solve the SSDD problem by representing documents as multisets and using the Jaccard Coefficient (JC) as a similarity measure. We first illustrate how the computation of Jaccard Coefficient can be reduced to the computation of intersection size between the multisets. Then, we propose a novel way to securely approximate the size of intersection between multisets using Bloom filters and hash functions, without significant reductions in security and accuracy. Our proposed protocol exploits a unique property of Bloom filters -that computing the dot product of two Bloom filters yields their intersection size.
DOI
10.1109/BDS/HPSC/IDS18.2018.00015
Montclair State University Digital Commons Citation
Forman, Sam and Samanthula, Bharath Kumar, "Secure Similar Document Detection: Optimized Computation using the Jaccard Coefficient" (2018). Department of Computer Science Faculty Scholarship and Creative Works. 532.
https://digitalcommons.montclair.edu/compusci-facpubs/532