Active noise correction graph embedding algorithm based on active learning

Abstract

The invention discloses an active noise correction graph embedding algorithm based on active learning, and belongs to the field of machine learning and data mining. Based on the idea of active learning, and a method of combining noise possibility and graph centrality is adopted. Under the condition that the label budget is given, more noise in the nodes with the mark graphs is selected through active learning. The structure of the graph is considered, and noise nodes with large information amount and representativeness are selected as much as possible. Compared with a traditional noise detection and active learning method, the invention has higher node classification accuracy. Not only can label nodes with noise be selected as much as possible, but also points with great influence on the structure of the graph can be found. Under the condition that labels requested to be labeled are the least, the performance of the classifier can be greatly improved.

Description

technical field [0001] The invention relates to the technical field of machine learning and data mining, in particular to an active noise correction graph embedding algorithm based on active learning. Background technique [0002] In various scenes in the real world, there are various network structures used to represent the interrelationships between objects. In computer science, network structures are represented as graph structures containing nodes and edges. However, most graph structures have the characteristics of large structure and large space overhead, so the computational task is heavy. Graph embedding is an effective way to solve this problem. Graph embedding converts graph structure information into low-dimensional dense real vectors, which are mapped to a low-dimensional latent space. And graph embedding can preserve the structural information and properties of the graph, and be used as the input of existing machine learning algorithms. Deep Walk method [1],...

AI Technical Summary

Problems solved by technology

[0003] However, the acquisition of labels in the real world is expensive, and it is generally difficult to obtain the labels of all nodes in the graph structure

In addition, the label data in the graph structure is not always correct, and there may be a certain amount of noise

Using these noisy node labels for training will affect the performance of graph embedding and classifiers, thereby reducing the accuracy of node classification

Images