Game-theoretic Privacy Preserving Constructions for Rational and Malicious Secret Sharing Models for Collaborative Frequent Itemset Mining

Abstract

Collaborative data mining amongst parties necessitates resolving the concern of maintaining the privacy of their data. Shamir's secret sharing, an information theoretically secure scheme, has been proposed for different scenarios in privacy preserving distributed frequent itemset mining (PPDFIM). However, one of the goals of PPDFIM is to ensure maximum participation from the contending participants. Hence, the scheme proposed must incorporate preventive and corrective measures, aimed at correcting the negatively performing participants. We resort to a game-theoretic approach for the purpose. We extend our proposed verifiable secret sharing (VSS)-based scheme for PPDFIM with a novel game-theoretic scheme called 'ratings-based game-theoretic' (RGT) scheme. In our improved scheme for PPDFIM, the parties with negative intent improve their behaviour tending the setup towards the coveted nash equilibrium (NE). We compare our proposed non-game-theoretic VSS-based scheme and the proposed RGT scheme theoretically and empirically in terms of utility of output and reduction in execution cost.