Graph neural network recommendation method for dissociating false correlation in multiple behaviors

The invention discloses a graph neural network recommendation method for dissociating pseudo-correlation in multiple behaviors. The recommendation method processes multi-behavior recommendation from a brand new perspective, namely dissociating pseudo-correlation information existing in auxiliary behaviors. Specifically, the model combines mutual information estimation and deentanglement representation to improve the recommendation performance of the multi-behavior recommendation method. A graph convolution recommendation task is used as a main recommendation task to perform preliminary modeling on interaction between users and projects, and then intention representation of each user under a specific behavior is accurately captured through a cross-behavior self-attention mechanism, so that a subsequent pseudo-correlation dissociation process is supervised. Through a dual mutual information boundary and an optimization method thereof, the model can dissociate pseudo-correlation information in auxiliary behavior