Graph classification method based on frequently dense pattern

Abstract

The invention discloses a graph classification method based on a frequently dense pattern. As a universal dataset structure, a graph can be used for expressing various complex relationships among data objects in a plurality of scientific applications. In the method disclosed by the invention, firstly, a new graph characteristic, i.e., the frequently dense pattern, is put forward, wherein the frequently dense pattern can keep weight information and a local topological structure in the graph, and is an ideal graph characteristic. Then, through a series of algorithms, an ordered pattern with discrimination is extracted from the graph, and the frequently dense pattern of the discrimination is taken as the characteristic. Finally, on the basis of a SVM (Support Vector Machine), a classifier used for graph classification is constructed. By use of the method disclosed by the invention, the graph classification can be efficiently and accurately realized.

Description

technical field [0001] The invention discloses a graph classification method based on frequent dense patterns, which involves neural image processing, social network, frequent item mining, classifier construction, etc., and aims to realize accurate and efficient classification of graph data. Background technique [0002] As a general data set structure, graph can be used in many problems to represent the complex structural relationship between data objects. For example, construct graph data based on neuroimaging, and then analyze and study the graph through complex network and other technologies, or use graph structure to represent the structure of compounds. At present, the graph classification problem mainly studies the two classification problems, that is, the positive class and the negative class. The main goal is to build a classification model to separate the two. In recent years, many kinds of graph features have been used for graph classification. For example, node...

AI Technical Summary

Problems solved by technology

However, thresholding will lose a lot of weight information in the loss map, which will affect the final classification results

Second, most graph features (node ​​degree, clustering coefficient, etc.) only consider the information of a single node, while ignoring the information between multiple nodes

Obviously, these two shortcomings will greatly affect the final classification performance

Images