The threat posed by phishing scams to Ethereum’s security has grown significantly with the advancement of blockchain technology. As a result, the detection of phishing scams has emerged as one of the most prominent research areas in the field of blockchain. Most existing studies represent transaction information as a static subgraph and employ random walks to extract potential user features. However, real-world graphs often exhibit dynamic behavior and evolve over time. To address these challenges, we introduce a novel approach called Dynamic Weighted Node Classification (DWNC). In this approach, we partition transaction records into multiple temporal snapshots based on time. We then capture the structural and temporal characteristics of the nodes using the structural aggregation module and the time aggregation module, respectively. Finally, we leverage the learned features for classification purposes. The proposed DWNC method demonstrates superior performance in classification, as evidenced by its evaluation on nine benchmark and Ethereum datasets.