Unveiling Hidden Communities: A Graph Clustering Approach to User Interactions and Closeness

Abstract

The growth of social networking sites (SNS) and the expansion of the web have facilitated easy communication among people on a single platform. A graph containing nodes and edges linking the nodes can be used to depict a social network. While the nodes represent the people or entities, the edges depict how these entities interact with one another. People who tend to associate with one another in social networks who have similar choices, tastes, and preferences form virtual clusters or communities. Finding these communities can be helpful for a variety of purposes, including locating a shared research area in cooperative networks, locating a user base for marketing and recommendation, and locating protein interaction networks in biological networks. This study presents a new way to locate communities that uses local knowledge and node space similarity. We use graph embedding to improve Community Discovery (CD) in social networks by combining eigenvector centrality and closeness measurements. Tests on six real-world datasets, including DBLP, Amazon, and Ego-Facebook, reveal that the suggested hybrid model does better than classic algorithms like Louvain, Walktrap, and Infomap. It gets a maximum NMI of 0.91 and a modularity of 0.86. These results show that the method is strong and can be used on a broad scale, making it a good way to find significant community structures in big networks.