Cloud-Assisted Verifiable Threshold PSI for Unbalanced Sets With Application to Financial Fraud Detection

Abstract

As a critical security function in the financial sector, fraud detection is commonly carried out with support from third-party risk control agencies. However, traditional detection methods risk leaking users’ private information. In this work, we implement financial fraud detection using threshold private set intersection (TPSI), enabling secure, cross-institutional sharing and matching of risk information while protecting user privacy. In addition, considering the imbalance in set sizes between the sender and receiver, we design two cloud server-assisted verifiable TPSI protocols that decouple the receiver’s computational overhead from the cardinality of the sender’s set, thereby significantly reducing the computational burden. Furthermore, to address the potential presence of malicious cloud server-assisted, we incorporate verification mechanisms to ensure the security of protocol execution and the correctness of results. Broadly speaking, we outsource the primary computational tasks of the receiver, including the oblivious key–value store (OKVS) decoding operation and secret polynomial reconstruction, to the cloud servers, which then interacts with the sender to perform calculations and, finally, returns the results to the receiver for verification. Moreover, to further improve the protocol’s computational efficiency, we employ cuckoo hashing and simple hashing to preprocess both parties’ sets. Finally, we conduct experimental and theoretical analyses of the security and efficiency of the two protocols. The results demonstrate that the single-server approach significantly enhances computational and communication efficiency while maintaining security, compared to the dual-server approach.