Graph node clustering method based on orthogonal robust non-negative matrix factorization

Abstract

The invention discloses a flow chart of a graph node clustering method based on orthogonal robust NMF, and the process of the flow chart is simplified as follows: step 1, formally representing graph data as G = (V, E); 2, constructing a graph data adjacency matrix A; 3, constructing a graph node clustering model based on orthogonal robust non-negative matrix factorization; 4, solving the graph node clustering model to obtain a W matrix and an H matrix; and step 5, obtaining a graph node clustering result according to the matrix H. According to the method, orthogonal constraints are applied tothe node clustering indication matrix, the sparsity of the matrix can be further improved, the node clustering affiliation relation is clearer, meanwhile, the robustness of a clustering model for noise is improved by designing a target function through L2 and 1 norms, and therefore the clustering quality of graph nodes can be finally improved. In addition, orthogonal constraints are converted intoregular terms of a clustering model, and a Lagrange multiplier method is introduced, so that the solving process of the model is greatly simplified.

Description

technical field [0001] The present invention relates to the technical field of graph data analysis and mining, and more specifically, to a graph node clustering method based on orthogonal robust non-negative matrix decomposition. Background technique [0002] Data that can be represented in a graph structure is called graph data. For example, social networks, paper co-authorship networks, and protein interaction networks are all typical graph data. Through the graph structure, the data objects in graph data can be represented as graph nodes, and the connections between data objects are represented as graph edges. Clustering analysis of graph nodes (that is, data objects in graph data) reveals clusters of closely connected member nodes, while nodes between different clusters are sparsely connected. Graph node clustering has good application value, for example, it can be applied to mining user communities of social networks, research teams of paper co-authorship networks, and...

AI Technical Summary

Problems solved by technology

[0003] At present, the problem of graph node clustering has attracted the attention of many researchers and some solutions have been proposed. Among them, the method based on nonnegative matrix factorization (NMF) has the advantages of interpretable clustering results, suitable for soft clustering and convenient integration and utilization. Advantages such as prior knowledge are becoming more and more popular, but this type of method still has the problem of low clustering quality, and further optimization and improvement are needed

Images