Improved Link Entropy with Dynamic Community Number Detection for Quantifying Significance of Edges in Complex Social Networks

Abstract

Discovering communities in complex networks is essential in performing analyses, such as dynamics of political fragmentation and echo chambers in social networks. In this work, we study the problem of quantifying the significance of edges in a complex network, and propose a significantly improved version of the Link Entropy method. Using Louvain, Leiden and Walktrap methods, our proposal detects the number of communities in each iteration on discovering the communities. Running experiments on various benchmark networks, we show that our proposed method outperforms the Link Entropy method in quantifying edge significance. Considering also the computational complexities and possible defects, we conclude that Leiden or Louvain algorithms are the best choice for community number detection in quantifying edge significance. We also discuss designing a new algorithm for not only discovering the number of communities, but also computing the community membership uncertainties.

Evolutionary Approach for Detecting Significant Edges in Social and Communication Networks

“…The first one is the Link Entropy (LE) developed by Qian et al [28] and demonstrating high efficiency. The second one is the most recent method, the Improved Link Entropy (ILE) developed by Lubashevskiy et al [33]. The third one is the Evolutionary Approach (EA) proposed in the given paper.…”

“…Our research belongs to the direction of network science that is aimed at elucidating the resilience of networks whose links can be destroyed. In the given case, detecting links whose role is crucial in maintaining the network connectivity becomes one of the main issues in this scope [15], [28], [29], [30], [31], [32], [33].…”

“…Chen et al [40] turned to the bridgeness to quantify the edge significance in maintaining the network connectivity; this index is determined explicitly by the local information of the network topology. Finally, there have been developed a family of novel methods based on the probability distributions divergence, namely, link entropy (LE) by Qian et al [28], deep link entropy (DLE) by Ozaydin and Ozaydin [31], and the improved link entropy (ILE) by Lubashevskiy et al [33] demonstrated their high efficiency.…”

“…The decrease rate in the size R gc of the largest connected component (normalized to the total number of nodes R gc ) caused by removing the most significant edges has been employed as the corresponding efficiency criterion. Indeed, the better the method of detecting edge criticality and, consequently, the more suitable the link ranking, the faster the decrease in R gc caused by removing the edges ordered downward according to their significance [15], [28], [30], [31], [32], [33].…”

“…To examine the efficiency of the proposed approach, three standard benchmark real-world networks are decomposed employing the EA, LE, and ILE methods. The choice of the LE and ILE algorithms as competitive methods is caused by that their comparison with algorithms turning to the other acknowledged indices of the edge significance (betweenness centrality, degree product, bridgeness, diffusion importance, topological overlap, and k-path edge centrality) has demonstrated the superiority of LE [28] and its descendant, ILE [33].…”