Node interaction prediction method, model, terminal and medium

The invention discloses a node interaction prediction method and model, a terminal and a medium, and the method comprises the steps: obtaining a historical neighbor node sequence of interaction of each node in a target node pair within a preset historical duration at a t moment based on a continuous time dynamic graph; extracting features of the historical neighbor node sequence to obtain original features; counting the interaction frequency of the nodes to obtain interaction frequency characteristics; aligning the original feature and the interaction frequency feature to obtain an alignment embedding representation, and performing frequency enhancement on the alignment embedding representation feature in a frequency domain to obtain a frequency domain fusion feature; and based on the frequency domain fusion features of all the nodes in the target node pair, predicting the interaction probability between the nodes in the target node pair at the t moment, and obtaining an interaction prediction result. Accordi