A fast adaptive nearest neighbor clustering method based on structured anchor graph

Abstract

The invention provides a fast adaptive nearest neighbor clustering method based on a structured anchor graph. First of all, using K-Means algorithm generates representative anchors; Then, the initialsimilarity matrix is constructed for the original data points and anchor points. Then, the similarity matrix is updated iteratively by using the structured anchor graph technique, and the nearest neighbor assignment is performed adaptively. Finally, the clustering results are obtained directly according to the connected components of the graph corresponding to the final similarity matrix. The method of the invention reduces the dependence of the large-scale spectral clustering task on the initial similarity matrix weight, and can quickly obtain high-quality clustering results by iteratively optimizing the anchor graph structure.

Description

technical field [0001] The invention belongs to the technical field of machine learning and data mining, and in particular relates to a fast self-adaptive nearest neighbor clustering method based on a structured anchor graph. Background technique [0002] With the explosive growth of smart devices and the popularity of the Internet and the Internet of Things, people's behavior data is being collected anytime and anywhere. As one of the most widely used techniques in exploratory data analysis, clustering, an unsupervised data learning method, is being increasingly applied to the preprocessing of large amounts of unlabeled data by academia and industry. From statistics, computer science, biology all the way to social science or psychology. In almost every scientific field that involves empirical data processing, people try to gain their intuition about the data by identifying combinations of "similar behavior" in the data. [0003] Spectral clustering algorithm is one of the...

AI Technical Summary

Problems solved by technology

However, the clustering results of these algorithms seem to be suboptimal, most likely because these algorithms are not designed to effectively learn data relationships

On the one hand, the performance of spectral analysis is limited by the quality of the similarity matrix

On the other hand, the solution matrix obtained from spectral analysis still needs K-means to transform it into discrete partitions, the sensitivity of K-means to initialization makes clustering performance unstable, and the separation of programs makes joint optimization difficult, let alone Said to achieve better performance

Images