METHOD FOR SELECTING FEATURE OF CLUSTER BY USING ENTROPY WEIGHTING AND SVD

摘要

PURPOSE: A cluster feature selection method is provided to select attributes representing a cluster based on a clustering result in a post-processing step in a clustering method using the entropy weighting and the SVD(Singular Value Decomposition) which can be applied in selecting a representative word group from a web document cluster or a representative goods group from a customer cluster in electronic commerce transaction data. CONSTITUTION: The method comprises several steps. High dimensional input data, represented by a matrix, gets decomposed into three matrixes and the three matrixes getting stored at corresponding files(S10). Attributes are selected among the decomposed two matrixes for constructing input data of a clustering algorithm, K-means, and a clustering process is performed(S20). An entropy weight of the attributes is obtained on a basis of a clustering result and an attribute frequency of original input data, a similarity between the cluster and the attributes is measured by using the object group information affiliated at the cluster, and a remaining decomposed matrix, and a representative attribute set of each cluster is determined(S30).