COMPUTER SCIENCE AND ENGINEERING, ONLINE SOCIAL NETWORKS, COMMUNITIES DETECTION

Abstract

This paper examines the performance of the community detection in Online Social Networks (OSNs) in order to gain a better insight into the user interactions, information flow, and structural organization. The study provides a comparison of three common algorithms, including Louvain, Label Propagation, and GirvanNewman, on real-life Facebook and Twitter data. The algorithms were tested on modularity, normalized mutual information (NMI) and conductance, and the visual analytics and composite scoring were used to support the evaluation after preprocessing the datasets into undirected and unweighted graphs. The other algorithm, the Louvain algorithm, performed the best with the highest modularity (0.842, 0.711) and NMI (88.3%, 82.6%) and the lowest conductance (0.151, 0.193), indicating high cohesion and separation of communities. Label Propagation was quicker, but less precise, and GirvanNewman was computationally demanding and less efficient in large networks. The results indicate that Louvain provides the most attractive and scalable community detection solution to a wide variety of OSN structures. The choice of an algorithm must be based on network topology and computational efficiency.