The Research On Dynamic And Abstract Clustering Method Of High Dimensional Sparse Data

With the development of the Database and Internet,we must deal with a lot of data, in order to explore the undiscovered and useful information, we need efficient methods to deal with these original data.Data Mining is the new data analysis technology which emerges and develops for this reason.Clustering analysis is an important part of Data Mining,with data mining is applied more and more widely in many fields,Clustering analysis also achieves increasing attention.Up to now,many clustering algorithms are available,these algorithms are applied in many relative fields,such as commercial market analysis, biologic research, image processing, pattern recognition, WEB searching etc.In this paper,some relative theories and research situation of data mining and clustering are introduced first,then the problems which need to be resolved of clustering are indicated,we analyse clustering methods which are Partition-Based method, Hierarchical-Based method, Density-Based method, Grid-Based method, Model-Based method, we also analyse some representative clustering algorithms. This paper mainly deal with the unconventional value in high dimensional sparse data clustering,because the traditional clustering algorithms are difficult to deal with the unconventional value in highdimensional sparse data clustering,so we can not get a high quality clustering.This paper proposes a new dynamic clustering method on the basis of two-value property and minimum cost spanning tree,this method can dynamically cluster the data according to different threshold,and it also can consider the weight of the attribute, so it can get a reasonable clustering.There are various relations exist in the nature, many of them are fuzzy,we need use fuzzy theory to describe them,so another dynamic clustering method on the basis of multiple similarity degree and fuzzy similar matrix is proposed,it also can dynamically cluster the data according to different threshold and consider the weight of the attribute when it is used in high dimensional sparse data clustering.These dynamic clustering methods are analysed with experiments in this paper,the results of the experiments show that these algorithms are effective and feasible.