An Empirical Study of Network Reduction: The Measurement and Comparison

Abstract

Abstract Large-scale networks face challenges for analysis and visualization in social network analysis due to their enormous size. Network reduction and clustering are essential techniques for large-scale networks. This study proposed an analytic framework that combines degree distribution, clustering coefficient distribution, KS-statistic, and normalized adjusted ratio sampling (NARS) to measure the social network dataset before and after reduction. The proposed NARS ensures that the network can obtain a fair share of nodes based on cluster size. The proposed framework aims to compare and investigate the effectiveness of network reduction and clustering. To evaluate the framework, 20 datasets of undirected networks were tested. Results show that the proposed framework is able to compare the reduced network to the original network. Based on the experimental results, random walk, one of the network reduction methods, and its improved version, induced subgraph random walk methods, perform equivalently although random walk can provide faster computational time.