This paper proposes a transaction tracking method based on dynamic graph link prediction to tackle the challenges posed by blockchain’s anonymity, which facilitates money laundering, fraud, and other illicit activities. By constructing a dynamic transaction graph and combining Temporal Graph Neural Networks with Transformer models, the method effectively predicts transaction links and uncovers potential transactional relationships. Experimental results show that the proposed method outperforms existing approaches in terms of Accuracy, F1-score, and AUC, providing significant support for blockchain transaction regulation and financial crime prevention.