By considering the evolution of networks over time and the rich semantic and structural
characteristics of heterogeneous networks, DyMDGCN model is proposed in this paper. The model
uses multi-channel partition to deal with heterogeneous networks, and then combines RNN and
time attention capture evolution mode to obtain the final node embedding vector from and apply
it to link prediction. In order to verify the effectiveness of this method, this paper selects two real
dynamic heterogeneous network data sets of Twitter and EComm for experiments, and compares
the AUC index with the traditional algorithm. The results show that this method can deal with heterogeneous network information and capture the dynamic evolution information of the network,
and has a certain improvement in accuracy compared with the traditional algorithm.