Network characterization method based on adaptive structure and position coding

Abstract

The invention relates to a network characterization method based on a self-adaptive structure and position coding, which comprises the following steps of: firstly, extracting feature information and structure information of nodes in an original network diagram, and obtaining two sub-network diagrams and corresponding adjacent matrixes from the original network diagram; secondly, initializing position codes of the original network diagram in a random walk mode; then, the feature matrix of the original network diagram and the adjacent matrixes of the two sub-network diagrams are input into two structure encoders respectively, and node-level representations of the two sub-network diagrams are obtained; then, enabling the initialized position codes to pass through a position encoder twice to obtain two position codes based on attention; and finally, splicing the node-level representations corresponding to the two sub-network diagrams and the attention-based position codes according to the dimension of 1, and mapping the node-level representations and the attention-based position codes into network representations of the original network diagram through a full connection layer. According to the method, the position code and the structure code of the network diagram are fused, so that the network representation can contain both the structure information and the position information.

Description

technical field [0001] The invention belongs to the technical field of self-supervised graph network representation, in particular to a network representation method based on adaptive structure and position coding. Background technique [0002] In recent years, deep learning has revolutionized many machine learning tasks, from image classification and video processing to speech recognition and natural language understanding. The data used in traditional machine learning is usually Euclidean spatial data with a regular spatial structure, while more and more non-Euclidean spatial data with mining significance, such as electronic transactions, recommender systems, etc. Machine learning algorithms present serious challenges. Network representation methods can model irregular and unordered non-Euclidean spatial data, and convert the vertices, edges or subgraphs of the graph into low-dimensional embeddings, thereby capturing the internal dependencies of the data and retaining imp...

AI Technical Summary

Problems solved by technology

Stacking multi-layer convolutional networks means that each node gathers the characteristics of multi-hop neighbor nodes, which will make the characteristics of all nodes tend to be consistent, which will lead to serious over-smoothing

Images