Semi-supervised symbol network embedding method and system based on improved graph convolutional network

Abstract

The invention provides a semi-supervised symbol network embedding method and system based on an improved graph convolutional network, and the method comprises the steps: importing relational network data, and forming a directed symbol network; calculating the adjacency matrix of the directed symbol network to obtain a propagation adjacency matrix, and activating the propagation adjacency matrix byusing a symbol function to obtain a directed activated propagation adjacency matrix; constructing a symbol Laplacian matrix, and applying the symbol Laplacian matrix by the graph convolution networkto realize improvement of the graph convolution network; using the improved graph convolution network to carry out convolution operation on an input adjacent matrix, and obtaining embedding results ofdifferent degrees are acquired, so that a prediction problem of network link information can be solved.

Description

technical field [0001] The disclosure belongs to the technical field of network data or information expression and display, and specifically relates to a semi-supervised symbolic network embedding method and system based on an improved graph convolutional network. Background technique [0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art. [0003] With the rapid development of social media and the gradual maturity of deep learning technology, network representation learning has become the focus of industry and academia. Network representation requires keeping the original topology and semantic information of the network unchanged while learning the low-dimensional latent representation of nodes. For example, in the comment trust network, if each user can be represented by a multi-dimensional vector, the information expression of the user on the network can be quantified, so...

AI Technical Summary

Problems solved by technology

[0004] However, according to the inventor's understanding, the current graph convolutional network only supports undirected unsigned networks, and cannot be directly applied to directed symbolic networks, that is, the original graph convolutional network uses the unsigned network Laplacian matrix to have The excellent properties of positive semi-definite, apply Fourier transform to realize the convolution operation of spectral domain graph

However, the directed symbolic network does not have this excellent property, and the original graph convolutional network cannot learn the negative relationship in the symbolic network, resulting in a serious imbalance in the final embedding results, which cannot effectively create potential value for related fields.

If you ignore the negative edges in the symbolic network and treat the symbolic network as an unsigned network, you will not be able to achieve satisfactory embedding results, and you will not be able to perform subsequent tasks: for example, you cannot effectively deal with the edges in the symbolic network. Direction and sign cannot solve the problem of symbol propagation in symbolic networks, and thus cannot realize the form of spectral domain convolution in directed symbolic networks

[0005] To sum up, there is still no effective solution to the series of problems existing in the application of popular graph convolutional network methods to symbolic networks.

Images