Feature Learning Method for Large-Scale Hybrid Graphs Based on Structural Semantic Fusion

Abstract

The present invention provides a large-scale hybrid graph feature learning method based on structure-semantic fusion, including: obtaining the training semantic label information set Strain, obtaining the node pair set Pe, Pe={(u,v)}, traversing the node pair (u, v) v); judge whether the traversal node pair (u, v) is completed; if it is judged that the traversal node pair (u, v) is not completed, then negatively sample the node u, and calculate the connected loss function and the disconnected loss function; if judged Knowing that node u is in Vtrain, calculate the semantic loss function according to Strain; update the initialization feature representation of node u, the initialization feature representation of node v, and the initialization feature representation of the node obtained by negative sampling; repeatedly judge and traverse node pairs (u, v) Whether to complete until the traversal of the node pair (u,v) is complete. The large-scale hybrid graph feature learning method based on structure-semantic fusion provided by the present invention corrects the feature representation of nodes according to the semantic label information, takes the semantic label information as a part of graph feature learning, and improves the quality of graph feature learning.

Description

technical field [0001] The invention relates to the technical field of computer data analysis, in particular to a large-scale hybrid graph feature learning method based on structural semantic fusion. Background technique [0002] A large amount of valuable information can be mined from the graph, such as which nodes have high similarity, which nodes form a community, and what potential connection relationships may exist. As an important technology in the field of graph data mining, graph feature learning provides a basis for applying machine learning algorithms on graph data. The goal of graph feature learning is to generate a feature vector for each node in the graph, which can be used as the input of the machine learning algorithm to obtain an analysis result or model that conforms to the graph characteristics. [0003] A variety of graph feature learning methods have been disclosed in the prior art, among which a large number of works have been proposed on how to maintai...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a large-scale hybrid graph feature learning method based on structural semantic fusion, which solves the problem that the graph feature learning method in the prior art cannot mine the information contained in the graph node labels, and the graph feature learning quality is low. technical problem

Images