To solve the problem of upstream and downlink interference in cellular networks, a graph convolutional neural networks-based novel fault diagnosis method for semi-supervised cellular networks is proposed. The method designed in this study uses extremal gradient enhancement to select the optimal feature subset, and uses graph convolutional neural network to extract the fault depth feature. At the same time, the knowledge data fusion technology is raised to expand the training data of the fault diagnosis model. This technology utilizes Naive Bayes models for pre-diagnosis and enhances graph convolutional neural networks to control the influence of pre-diagnosis outcomes and training dataset size on model training accuracy. In the experiment, the fault diagnosis accuracy and efficiency of the raised method are better than those of the traditional network fault diagnosis methods. This algorithm can diagnose faults in complex cellular network environment, which has high accuracy and practicability, and can effectively improve user experience.