Fuzzy Aggregation and Label Propagation Based Social Community Detection Using Cuckoo Search

Abstract

A node in a social network is not part of just a cohesive group; it may be classified in many close or different communities. The community detection problem in social networks is similar to the clustering in networks where information is stored in node attributes and network structures. The Node-based features are often used in unsupervised algorithms that partially detect the local and overlapping communities. Networks displaying a community structure may exhibit hierarchical communities as well. Identification of hierarchical communities with community structure is a challenging task. This paper presents a two-step framework, i.e., aggregation and label propagation, to identify crisp and non-overlapping communities. Aggregation is an expansion-dissolution technique that results in crisp communities that don't require any prior information. We initially estimated random communities for each node and applied aggregation to improve them. The second step is the label propagation-based objective maximization method that takes improved crisp communities as fuzzy matrices and results in non-overlapping communities. The label propagation method is a label update mechanism for nodes with neighbors in a different community. The Label propagation function is maximized using cuckoo search and reported optimized communities. The two-step framework is empirically tested on real and simulated social networks. A comparative and contrast study is performed using performance and accuracy-based metrics to validate the framework and is found to be state-of-the-art.