False user identification method based on improved GraphSAGE algorithm

The invention discloses a false user identification method based on an improved GraphSAGE algorithm. The method comprises the following steps: calculating a collusion fraud degree between users based on score deviation and a score time interval in score data, and a comment text length and comment text similarity in comment data; constructing a user relation graph by using a user article interaction relation existing in the scoring data; extracting user scoring features from the scoring data by using a collaborative denoising auto-encoder; using the user score features as user initial features, using an improved graph neural network GraphSAGE algorithm to aggregate and extract center node user features from the user-user relation graph, and using a sigmoid activation function to calculate the classification probability of center node users based on the extracted center node user features, and performing supervised training of a false user identification model by using cross entropy loss as an optimization targ