Akademik Veri Yönetim Sistemi
NEURAL COMPUTING AND APPLICATIONS, cilt.1, ss.1-15, 2023 (SCI-Expanded)
Many algorithms have been developed to detect communities in networks. The success of these developed algorithms varies according to the types of networks. A community detection algorithm cannot always guarantee the best results on all networks. The most important reason for this is the approach algorithms follow when dividing any network into communities (sub-networks). The modularity of the network determines the quality of communities in networks. It is concluded that networks with high modularity values are divided into more successful communities (clusters, sub-networks). This study proposes a modularity optimization algorithm to increase clustering success in any network without being dependent on any community detection algorithm. The basic approach of the proposed algorithm is to transfer nodes at the community boundary to neighboring communities if they meet the specified conditions. The method called KO (Karcı–Oztemiz) optimization algorithm maximizes the modularity value of any community detection algorithm in the best case, while it does not change the modularity value in the worst case. For the KO algorithm’s test, in this study, Walktrap, Cluster Edge Betweenness, Label Propagation, Fast Greedy, and Leading Eigenvector community detection algorithms have been applied on three popular networks that were unweighted and undirected previously used in the literature. The community structures created by five community detection algorithms were optimized via the KO algorithm and the success of the proposed method was analyzed. When the results are examined, the modularity values of the community detection algorithms applied on the three different networks have increased at varying rates (0%, …,14.73%)