A Study on Cluster Partition of Hierarchical Clustering for Network Data

Abstract

Network data is data that expresses connections between objects. Network data consists of nodes representing objects and edges representing relationships between nodes. Individuals in SNS, hypertext links on the World Wide Web, and e-mail and message exchanges are represented as network data. In recent years, the analysis of large-scale network data has been attracting attention. Clustering is one of the data mining tools for processing large-scale data. Clustering is a method for dividing a set of objects into subsets called clusters. The Louvain method is known as a clustering method for analyzing network data. The method joins clusters sequentially. Hierarchical clustering is known as a clustering method that joins clusters sequentially. The Louvain method can be regarded as hierarchical clustering from a clustering perspective. In this study, we focused on the similarity between the cluster partition generated by the Louvain method and hierarchical clustering. We conducted experiments to compare the two methods. In the experiment, the results were evaluated by ARI and Modularity, and the similarity between the two methods was visualized by MDS. It was suggested that the Ward method shows high Modularity and similarity with the Louvain method from the experimental results.