Adjacency matrix-based graph feature extraction system, graph classification system and methods

Abstract

The present invention provides an adjacency matrix-based graph feature extraction system, an adjacency matrix-based graph feature extraction picture classification system and corresponding methods. According to the systems and methods, connection information elements in an adjacency matrix corresponding to a graph are concentrated at a specific diagonal region of the adjacency matrix, and non-connection information elements are reduced in advance; a filtering matrix is used to extract the subgraph structure of the graph along a diagonal direction; a stack convolutional neural network is used to extract a larger subgraph structure; and therefore, computational complexity and computational quantity are greatly reduced, the restrictions of computational complexity and the restrictions of window size can be eliminated; and a large multi-vertex subgraph structure and the deep features of implicit correlation structures from vertices and edges can be captured through a small window, and theaccuracy and speed of graph classification can be improved.

Description

technical field [0001] The invention belongs to the field of artificial intelligence, and in particular relates to a graph feature extraction system based on an adjacency matrix, a graph classification system and a method. Background technique [0002] A graph in graph theory is a graph composed of a number of given points and a line connecting two points. This graph is usually used to describe a certain relationship between certain things, using points to represent things, and using A line connecting two points indicates that there is a certain relationship between the corresponding two things. A graph (Graph) G in graph theory is an ordered binary (V, E), where V is called a vertex set (vertex set), that is, a set composed of all vertices in the graph, and E is called an edge set (edge ​​set). ), which is the set of edges between all vertices. Simply put, vertices represent things, and edges represent relationships between things. Graph is a kind of non-grid data. The c...

AI Technical Summary

Problems solved by technology

However, it still has some disadvantages

First of all, the selection of w central vertices will limit the number of subgraphs, so it is impossible to guarantee that all subgraph structures can be extracted; secondly, the PSCN method is still limited by the window size, and the selection of the neighborhood is from a window size smaller than 10 k decision, because a larger window size k will result in unacceptable time-consuming and memory usage; again, PSCN is not effective for deep learning when using a smaller window size k, because when the input graph has more than the default window size It loses the complex subgraph features when densely connected features of , and PSCN's classification results are sensitive to the labeling process, which is to sort the vertices in the domain; so their labeling method is suitable for a data set , but may fail on another dataset

[0012] To sum up, there are two main problems in the classification of graphs in the prior art methods: one is that when analyzing a graph (graph) as a whole object, it is impossible to select features that can contain both display topological information and deep hidden information. Represent graphs; secondly, when subgraphs are used as the features of graphs, the size of subgraphs is subject to the selection of window size (windowsize) k, which makes it difficult to capture large and complex subgraphs, making the classification of graphs Not very accurate

However, the methods in the prior art cannot effectively extract the larger subgraph structure in the graph, and thus cannot perform a good feature representation on the graph.

Images