Blockchain address risk behavior identification algorithm based on neural differential equations

First, the Tgm-ODE model was proposed, which realized the identification of criminal behavior using USDT for wallet addresses on the wavefield chain. Then a neural ordinary differential equation model (Neural ODE) was used to learn the continuous changes in the characteristics of node addresses with different transaction time intervals. At the same time, a gate mechanism was introduced to filter out the impact of neighboring node addresses on the central node. The gate mechanism design incorporated the strength of transaction correlation between node addresses. Finally, the self attention mechanism was used to fuse the node address features at different transaction times, outputting the feature representation of node addresses. Experimental results show that the Tgm-ODE model can effectively capture the dynamic changes of node addresses with irregular transaction intervals, and outperforms traditional detection models in terms of precision, recall, and F1 metrics in the test set.