A visual analysis method of graph neural network based on force map

Abstract

The invention discloses a visual analysis method of a Graph Neural Network (GNN for short) based on a force-guided graph. The method includes the following steps: S1, constructing a graph neural network model, and statistically calculating the middle hidden layer parameters of the graph neural network or the middle The output of the hidden layer; S2. Construct the force map model, and use the parameters of the middle hidden layer of the graph network neural network or the output of the middle hidden layer as the input of the force map model; S3. According to the force situation of the nodes in the force map, iterate Update the position of the nodes in the force map, and obtain the final layout when all the nodes in the graph are balanced or the update displacement is less than the threshold; the system of the present invention corresponds to the method; the present invention observes the update process of the graph neural network parameters from the perspective of visualization, and then explains The learning effectiveness of graph neural network enhances the interpretability of graph neural network.

Description

technical field [0001] The invention relates to the technical field of visual analysis, in particular to a visual analysis method of a graph neural network based on a force map. Background technique [0002] With the rapid growth of data scale and computing power, deep learning technology represented by convolutional neural networks (CNN) and recurrent neural networks (RNN) has been widely used in the industry and has produced huge dividend. Compared with traditional machine learning methods, deep learning has stronger feature expression and learning capabilities: CNN has achieved great success in the field of computer vision. The effect on the data set is better than that of traditional machine learning methods; RNN has also made great breakthroughs in the field of natural language, and has made breakthroughs in language recognition, machine translation, text classification and other issues. [0003] But in the real world, in addition to regular grid data and time series ...

AI Technical Summary

Problems solved by technology

In 2016, Defferrard et al. proposed an improved graph convolutional neural network (Graphconvolutional neural networks, GCNN) for the high computational complexity of the graph neural network proposed by Henaff et al., which is approximated by Chebyshev polynomials. matrix operations, resulting in approximately smooth filters in the spectral domain

[0005] The graph convolutional neural network, like other deep learning models, lacks a rigorous theoretical basis for mathematical proof, and its interpretability is not as good as traditional machine learning.

Images