Network Link Connectivity Prediction Based on GCN and Differentiable Pooling Model

Graph Neural Networks (GNN) are widely used in node classification and link prediction, and have achieved very good results. However, the current GNN does not have a deep hierarchical representation of learning graphs, and is mostly a simple weighted superposition. This paper proposes a link predict...

Full description

Saved in:
Bibliographic Details
Main Authors: Kuang, Naixue, Zuo, Yanzhi, Huo, Yonghua, Jiao, Libin, Gong, Xingle, Yang, Yang
Format: Conference Proceeding
Language:English
Subjects:
Complexity theory
Data mining
differentiable pooling
Feature extraction
graph convolution neural network
Labeling
link prediction
node labeling
Prediction algorithms
Predictive models
Simulation
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Graph Neural Networks (GNN) are widely used in node classification and link prediction, and have achieved very good results. However, the current GNN does not have a deep hierarchical representation of learning graphs, and is mostly a simple weighted superposition. This paper proposes a link prediction algorithm based on graph Convolutional Neural Network (GCN) and an improved differentiable pooling model, which combines GCN and differentiable pooling to extract deep-level graph features. At the same time, the Principal Component Analysis (PCA) dimensionality reduction method is used to improve the assignment matrix generation method is proposed to improve link prediction accuracy; an improved dual distance node labeling is proposed to mark nodes to construct a node information matrix; a common neighbor method with pruning is proposed to extract subgraphs while reducing link data the complexity. The simulation results show that the link prediction algorithm based on the graph convolutional neural network and the improved differentiable pooling model performs better in link prediction.
ISSN:2770-1603
DOI:10.1109/ICAIT56197.2022.9862715