According to CoinMeta, as reported by the Nanhua Morning Post, researchers from the People's Public Security University of China have developed a framework for identifying illegal cryptocurrency transactions. This framework combines dynamic graph neural networks, historical pattern memory modules, and large language models. On the publicly available Bitcoin transaction dataset, the model achieved an overall accuracy of 89.4%, with an 89.1% precision rate in identifying illegal transactions and a recall rate of 64.5%. The model is also capable of generating risk scores, key evidence, and natural language analysis to assist in tracking evolving online criminal patterns such as money laundering.