Filtering-based graph clustering method

Abstract

The invention provides a filtering-based graph clustering method, which comprises the following steps of: firstly, carrying out filtering operation on original data characteristics by using a low-passfilter, and then, improving an optimization objective function based on a subspace clustering model, so that the optimization objective function excavates structural information and characteristic information of a graph at the same time, and calculates an affinity matrix; and finally, enabling the affinity matrix to be symmetrical, and obtaining a graph clustering result through spectral clustering. The method is very universal, a concise and efficient graph clustering algorithm model is provided based on the filtering technology, and compared with a method based on deep learning, the calculation process of a large number of parameters is avoided. Compared with the existing method, the method disclosed by the invention shows huge advantages on a plurality of widely applied data sets.

Description

technical field [0001] The invention belongs to the field of attribute graph clustering, in particular to a filter-based graph clustering method. Background technique [0002] Graph clustering is a long-standing problem in machine learning, data mining, and pattern recognition, which has countless application scenarios, such as community analysis, protein structure analysis, etc. The input data of graph clustering is generally an attribute graph, and the output result is the category of each node. The attribute graph is composed of each node's own characteristics and the edge set between nodes, which is a good representation method of Non-Euclidean Structure Data. Since graph clustering is an unsupervised learning task, it is very difficult to achieve good results on data. Its performance is easily affected by many factors, such as clustering methods, data noise, etc. Various graph clustering techniques, such as K-means and spectral clustering, are particularly popular. ...

AI Technical Summary

Problems solved by technology

However, the original subspace clustering method does not fully consider the characteristics of the data.

Generally speaking, there is noise in the data, which will have a great impact on the clustering results.

At the same time, the original subspace clustering method only mines the characteristic information of the graph, does not consider the structural information of the graph, and does not fully mine the known information of the graph

Images