Node classification method based on dual-channel knowledge distillation

Abstract

The invention relates to the technical field of semi-supervised node classification, in particular to a node classification method based on dual-channel knowledge distillation, and the method comprises the steps that a dual-channel teacher model and a corresponding student model are set; during training, the two-channel teacher model learns knowledge information of a spatial topological structure and knowledge information of node and neighbor node feature attributes, and instructs the student model to train based on knowledge levels corresponding to the two kinds of knowledge information, so that the student model keeps structure-based priori knowledge and feature-based priori knowledge; during testing, the trained student model is used for node classification. According to the node classification method based on dual-channel knowledge distillation, the knowledge comprehensiveness and classification accuracy of the student model can be improved, so that the classification effect of node classification is assisted to be improved.

Description

technical field [0001] The invention relates to the technical field of semi-supervised node classification, in particular to a node classification method based on dual-channel knowledge distillation. Background technique [0002] Graphs are basic data structures that describe pairwise relationships between entities, such as social networks, academic networks, and protein networks. Learning and mining graph data can help solve various practical application problems. Among them, the node classification of semi-supervised learning is an important task of graph data mining. It predicts other nodes in the graph by giving the labels of a small number of nodes in the graph. The label of the node. Graph Convolutional Network (Graph Convolutional Network, GCN) generates a new representation of nodes by aggregating the features of neighboring nodes, and extracts effective features from graph data. In recent years, it has been widely used in many fields, including node classification,...

AI Technical Summary

Problems solved by technology

However, the applicant found that in the above-mentioned existing node classification methods, the knowledge distillation framework only focuses on one-sided knowledge guidance, while ignoring other levels of knowledge, so that the student model cannot fully learn knowledge information from the teacher model, which leads to student The knowledge of the model is not comprehensive

At the same time, the existing knowledge distillation framework cannot make good use of the homogeneity phenomenon, because it ignores the prior knowledge in the original data set, resulting in low classification accuracy when classifying student model nodes

Images