Analysis of Information Propagation in Ethereum Network Using Combined Graph Attention Network and Reinforcement Learning to Optimize Network Efficiency and Scalability

Blockchain technology has revolutionized the way information is propagated in decentralized networks. Ethereum, as a major blockchain platform, plays a pivotal role in facilitating smart contracts and decentralized applications. Modeling information propagation dynamics in Ethereum is crucial for ensuring network efficiency, security, and scalability. In this study, we introduce three innovative theorems, aiming to use Graph Attention Network (GAT) to analyze the information propagation patterns; while our major contribution is to develop a combined GAT and Reinforcement Learning (RL) method to enhance the network efficiency and scalability by optimizing the gas limits for block processing. It learns the best actions to take in various network states, ultimately leading to improved Ethereum network efficiency and throughput and optimize gas limits for block processing. Additionally, we explore methods for effectively aggregating transaction data by capturing graph structures and updating node embeddings for transaction pattern prediction. To evaluate scalability, we implement and compare three Graph Neural Network (GNN) models---GraphConv, GraphSAGE, and GAT---comparing their performance at scale.