Clustering Ensemble for Categorical Geological Text Based on Diversity and Quality

Abstract

Clustering analysis for geological text makes the navigation, retrieval or extraction of geological text more effectively. Clustering ensemble can be employed to obtain more robust clustering results. However, most generation approaches focus on the diversity of clustering members rather than their quality. Too much emphasis on the diversity of clustering members reduces the accuracy of clustering results. In order to solve the problem, a new generation method of clustering members is proposed in this paper. Hierarchical clustering algorithm and k-means algorithm alternately combined with random projection method are employed to generate diverse base members and a new selection strategy for the number of clusters is presented to improve the quality of clustering members. Furthermore, a clustering ensemble framework for geological text is constructed. The framework involves geological text preprocessing, geological text feature representation, clustering members generation and ensemble integration. Experimental results on two UCI datasets and one real-world geological text demonstrate that the clustering ensemble based on diversity and quality is superior to those clustering ensemble algorithms that only consider the diversity of clustering members.