With its decentralized, tamper proof, transparent and traceable characteristics, blockchain technology has shown great potential in fields such as finance, supply chain, and the Internet of Things. However, the public transparency of its ledger poses a serious challenge to user transaction privacy. Traditional privacy protection schemes such as homomorphic encryption and zero knowledge proofs can enhance privacy, but often struggle to balance computational overhead, communication costs, and data availability. This article explores the innovative application of neural networks in blockchain privacy protection and proposes a transaction obfuscation model based on generative adversarial networks. This model utilizes a generator to learn the statistical features of raw transactions and generate difficult to track obfuscated transactions, while ensuring the validity and compliance of obfuscated transactions through discriminators and blockchain verification contracts. The experimental results show that compared with traditional obfuscation methods and differential privacy methods, the proposed model significantly reduces the consumption of privacy budget and computation delay while ensuring high transaction utility (such as reducing address correlation by more than 85%), achieving a better balance between privacy and utility. This study provides new ideas for building efficient and practical blockchain privacy enhancement solutions.
Gao et al. (Thu,) studied this question.