Multi-stage Hierarchical Clustering Method Based on Hypergraph

Abstract

Clustering analysis is a data analysis technique, it groups a set of data points into multiple clusters with similar data points. However, clustering of high dimensional data is still a difficult task. In order to facilitate this task, people usually use hypergraphs to represent the complex relationships between high dimensional data. In this paper, the hypergraph is used to improve the representation of the complex high dimensional data, and a multi-stage hierarchical clustering method based on hypergraph partition and Chameleon algorithm is proposed. The proposed method constructs a hypergraph in the shared-nearest-neighbor (SNN) graph from the dataset and then employs a hypergraph partitioning method hMETIS to obtain a series of subgraphs, finally those subgraphs are merged to get the final clusters. Experiments show that the proposed method is better than Chameleon algorithm and the other four clustering methods when applied on four UCI datasets.