Community detection in social networks through similarity virtual networks

Abstract

Smart marketing models could utilize communities within the social Web to target advertisements. However, providing accurate community partitions in a reasonable time is challenging for current online large-scale social networks. In this paper, we propose an approach to enhance community detection in online social networks using node similarity techniques. We apply these techniques on unweighted social networks to detect community structure. Our proposed approach creates a virtual network based on the original social network. Virtual edges are added during this pre-processing step based on nodes' similarity in the original social network. Hence, a virtual link is established between any two similar nodes. Then the landmark CNM algorithm is applied on the generated virtual network to detect communities. This approach, labelled Similarity-CNM is expected to further maximize the quality of the inferred communities in terms of modularity and detection speed. Our experimental evaluation study asserts these gains, which accuracy is supported by a study based on Normalized Mutual Information Measure to determine how similar are the actual communities in the original network and the ones found by the proposed approach in this paper.