A one-step spectral clustering method based on spectral rotation

Abstract

The invention discloses a one-step spectral clustering method based on spectral rotation, which relates to the field of computer big data information technology, and solves the technical problem thata spectral clustering method with simplified steps of spectral clustering and high clustering accuracy is provided. Means clustering optimization and transformation matrix learning are integrated intoa framework, using the reduced dimension feature space of the original dataset to learn the relationship matrix by analyzing the original k-Means result to find the better clustering partition hyperplane, and get the better clustering partition result. The invention simplifies the spectral clustering step, the clustering time complexity of the large data is linear, and the coding only relates toa simple mathematical model, which is easy to implement and has high clustering accuracy.

Description

technical field [0001] The invention relates to the technical field of computer big data information, in particular to a one-step spectrum clustering method based on spectrum rotation. Background technique [0002] With the rapid development of the Internet, especially the mobile Internet, a large amount of data is continuously collected and organized. The current main research on big data knowledge discovery includes four aspects: division, clustering, retrieval, and incremental learning. Clustering has become a research hotspot because it can help discover hidden information in big data. [0003] Among many clustering methods, spectral clustering has become a hot research direction because it can cluster on the sample space of any shape and converge to the global optimal solution. The prior art spectral clustering method is usually divided into three steps, first is the construction of the relationship matrix, then the learning of the spectral representation, and finally...

AI Technical Summary

Problems solved by technology

For the spectral clustering method of the prior art, it is an important step to construct a new reliable high-quality relation matrix, and the relation matrix constructed by the spectral clustering method of the prior art is obtained from the original Euclidean feature space , cannot accurately reflect the real relationship between the data, and then the subsequent processing of this relationship matrix cannot obtain accurate subspace division

In addition, the division plane selected for the final clustering division using k-means is not a better division plane for the real data set distribution, so it has a greater impact on the clustering accuracy

Images