Node processing method and device based on graph neural network

The embodiment of the invention provides a node processing method and device based on a graph neural network, and the method comprises the steps: inputting a target heterogeneous graph into the graph neural network, determining an initial node of the target heterogeneous graph in the graph neural network, at least two associated nodes corresponding to the initial node, and at least two connection edges between the initial node and the at least two associated nodes; determining initial node information of the initial node, associated node information of the at least two associated nodes and side information of the at least two connecting sides; according to a preset aggregation algorithm, aggregating the association node information of the at least two association nodes to obtain aggregated node information, and aggregating the edge information of the at least two connection edges to obtain aggregated edge information; and updating the initial node information according to the aggregation node information and