Multi-view map clustering algorithm based on tensor singular value decomposition

Abstract

The invention provides a multi-view map clustering algorithm based on tensor singular value decomposition. The algorithm expresses a probability transfer matrix of all view data by using a three-ordertensor X. The tension has low rank in horizontal, longitudinal and vertical directions, so that the algorithm provided by the invention adopts a multi-rank based on tensor singular value decomposition (Tensor-SVD) to represent the low rank of the tension in each dimension. The Tensor-SVD is generated based on tube convolution, so that the Tensor-SVD can express correlation in a spatial structuremore fully than the other tension decomposition modes and a method of modeling based on a two-dimensional structural relationship, fast computation can be performed via Fourier transform, and the computation efficiency is improved. Therefore, modeling is performed based on Tensor-SVD, so as to be more scientific, faster and more efficient, and an experimental result indicates that the multi-view clustering effect can be effectively improved.

Description

technical field [0001] The present invention relates to the field of data mining, more specifically, to a multi-view spectrum clustering algorithm based on tensor singular value decomposition. Background technique [0002] The multi-view clustering problem mainly improves the performance of clustering by integrating useful complementary information in multiple views. At present, related algorithm research can be roughly divided into three categories: based on multi-image fusion algorithm, based on collaborative training algorithm, and based on subspace learning algorithm. [0003] First, based on the multi-image fusion algorithm. The idea of ​​this type of method is to construct a graph structure for each view, and then fuse these graph structures. In 2007, the research of Professor Dengyong Zhou and Professor Christopher J.C.Burges of Microsoft Research showed that a random walk was defined for each graph structure, and then a Markov mixture model was defined for all rand...

AI Technical Summary

Problems solved by technology

[0006] At present, the existing methods of multi-view clustering based on multi-kernel learning all use matrix to represent each view data, and only use two-dimensional relational structure to model and solve to obtain the fusion view, ignoring the overall spatial structure information between views, resulting in Insufficient use of effective information, clustering effect is not good enough

Images