Semi-supervised hierarchical clustering method based on ultrametric distance matrix

Abstract

The invention provides a semi-supervised hierarchical clustering method based on an ultrametric distance matrix. The method comprises steps as follows: step 1, a closed convex set of inequality constraint is defined, and parameter estimation is projected to the closed convex set; step 2, an estimation solution vector is updated by reducing a variable vector formed in the projection; and step 3, iteration projection is performed until a given constraint fixed set is converged to the least-square optimal solution. According to the method, a semi-supervised hierarchical clustering frame based on an ultrametric tree diagram distance is taken as a research background, an optimized method is adopted, and the semi-supervised hierarchical clustering method based on the ultrametric distance matrix is provided and used for improving the efficiency and accuracy for solving the semi-supervised hierarchical clustering problem.

Description

technical field [0001] The invention belongs to the clustering technology in data mining, in particular to a semi-supervised hierarchical clustering method based on an ultra-metric distance matrix realized by an optimization technology. Background technique [0002] The process of grouping a collection of physical or abstract objects into similar object classes is called clustering. Clustering problems arise in many disciplines and are widely used. Basically, the purpose of clustering is to classify given samples into corresponding clusters so that samples in the same cluster are similar to each other and samples in different clusters are different from each other. Based on the way clusters are generated, clustering methods can be divided into two categories: partitional clustering and hierarchical clustering. Partitioning clustering generally decomposes a data set into some disjoint clusters, and this decomposition is usually optimal in terms of some pre-defined objective...

AI Technical Summary

Problems solved by technology

[0006] The HAC algorithm is very simple in the cluster object, it can use a similar method to find clusters of different shapes, but HAC also has some disadvantages: (1) HAC has a high time complexity, for example, for the centroid point algorithm (priority queue method), its time complexity is O(N2logN); (2) The effectiveness of obtaining clusters with pedigree graphs is limited

Many efficiency methods exhibit shifting patterns to the lower layers of the pedigree graph, which can lead to imprecise estimates of the optimal number of clusters

Images