Abstract: The dissimilarity measure and clustering approach about the heterogeneous dataset were studied, and a struclure-based entropy clustering SEC algorithm was presented in this paper. Data often do appear in homogeneous groups,the SEC utilizes these structural information to improve the clustering accuracy. Unlike the distribution of numeric data,nominal data are often unbalancedly distributed,whose distribution are often unrelated with their distance measure,due to the above, a new structural information-based entropy computing technology was proposed. By mining the clues in structural information, constructing the weight implying the different distribution information of nominal and numeric attributes, the SEC can automatically identifies the initial locations and number of cluster centriods, and exhibits its robustness to initialization and no iteration in algorithm. Experimental results comparing with other references demonstrate that the proposed method has promising performance.

Key words: Heterogeneous data, Dissimilarity measure, Clustering clue, Structural entropy, Clustering algorithm