Clustering method for performing matrix decomposition by taking subset grouping as auxiliary information

Abstract

The invention relates to a clustering method for performing matrix decomposition by taking subset grouping as auxiliary information, which is characterized in that grouping information of a certain number of subsets is obtained through collecting grouping results of a user for different subset to act as guidance for clustering, a close neighbor set and a far neighbor set of objects in the subset are obtained based on the results, and the category of each object in the subset is enabled to be close to the category of objects in the close neighbor set but be different from the category of objects in the far neighbor set through a mode of adding regular items to an objective function for matrix decomposition so as to complete clustering. The clustering method not only considers the decomposition error, but also reduces the difference between grouping of objects in the subset and grouping of objects in the close neighbor set and increases the difference between grouping of objects in the subset and grouping of objects in the far neighbor set at the same time in matrix decomposition based on a grouping result of the subset in the previous step, thereby achieving satisfactory results without the need of excessive number of constraints, and is fast in clustering, high in efficiency and low in labor cost in practical application.

Description

technical field [0001] The invention belongs to the technical field of electrical digital data processing, and in particular relates to a clustering method based on machine learning and data mining and using subset grouping as auxiliary information for matrix decomposition. Background technique [0002] In data processing and analysis problems in many fields, it is necessary to use clustering algorithms to group samples in a data set, and then quickly browse, analyze and process the internal structure of the entire data set based on the grouping results. [0003] In order to improve the accuracy of traditional unsupervised clustering methods, semi-supervised clustering using a small amount of supervised information is proposed. Most of the existing semi-supervised clustering methods use the form of constraints between two objects to achieve supervision, so that two objects related to the "must link" constraint are classified into the same class, while the "cannot link" The ...

AI Technical Summary

Problems solved by technology

[0005] The technical problem solved by the present invention is that in the prior art, the semi-supervised clustering method has a total of n objects for a data set containing n objects 2 There are two possible pairwise relationships, so a large enough number of constraints is required to achieve satisfactory results. In practical applications, obtaining a large number of constraints requires a large amount of human labor and low efficiency, thus providing an optimized A Clustering Method Using Subset Grouping as Auxiliary Information for Matrix Decomposition

Images