Software defect prediction method based on graph convolutional neural network

The invention discloses a software defect prediction method based on a graph convolutional neural network. Defect types of input code files are predicted by utilizing a GCN algorithm training model. According to the method, feature extraction is carried out on the source code file of the software through the Bert model; association between files in the source code is realized by constructing an abstract syntax tree; and then files possibly having defect transmission in the codes are associated by using an association algorithm Apriori, and finally, the association relationship between the feature vectors of the source files is used as an adjacent matrix to be used as input, so that the training of the GCN model is realized. When whether the software code file has defects or not is judged,the code file is automatically converted into the feature vector corresponding to the code file as the input of the model, and the GCN model outputs the code file which may have defects, so that the workload of testers is gre