Efficient Privacy-Preserving Truth Discovery and Copy Detection in Crowdsourcing

Xiu Susie Fang, Xinyang Du, Hao Chen, Ziqi Wei, Yong Zhan, Guohao Sun

Abstract: Researchers continue to focus on privacy-preserving truth discovery and achieve certain results with the increasingly popular trend of privacy protection. However, the common existence of copiers among workers is overlooked in existing privacy-preserving truth discovery, which causes decreased accuracy. Since methods based on encryption or perturbation may easily introduce noise and lose correlation between original data, it is challenging to detect copiers on privacy-preserving data. To address this challenge, in this paper, we propose an anti-copy iterative model based on lightweight homomorphic encryption, called CAPP-TD. First, we propose a lightweight privacy protection mechanism based on Paillier homomorphic encryption that preserves the correlation of privacy data. Compared with traditional homomorphic encryption-based algorithms, it requires less communication and computation overhead to perform truth discovery with copy detection. We then propose an iterative truth discovery method that can efficiently detect copy relationships in encrypted data and exclude copiers from truth inference to improve accuracy. Experimental results on both real-world and synthetic datasets and thorough security analysis demonstrate that CAPP-TD protects crowdsourcing systems from adversaries and enables highly accurate truth discovery.