DFCD: Density-Based Fuzzy Community Detection Approach With Biased Random Walk Guided Node Sampling

Research on overlapping community detection is still significant due to the rapid growth of different types of complex networks. Among the types of community detection, density-based clustering approaches did not get much attention in the field of overlapping community detection. We found two main drawbacks of the density-based approaches: 1) lack of research in computing the proper long-range distance (up to several hops) among the nodes in graph dataset; and 2) difficulty in computing overlapping belonging coefficients of the nodes in terms of diverse and irregular shaped clusters of graph datasets where distances among few nodes are available only (even not the positions of the nodes). In this article, we propose a novel overlapping community detection approach which we term as “density-based fuzzy community detection (DFCD)”, based on local densities of the nodes. Here, we capture the network topology in the form of bulk node-streams (node sampling) by employing biased random walk. We propose a novel way to compute long range distances among the nodes by utilizing the nodes-streams to find fixed number of global neighbors for each node, and accordingly we compute local densities of the nodes. We propose three relations between nodes and communities, where local densities and distances among the nodes are known only. Finally, we propose a novel belonging computation function based on the three relations, local densities, distances among the nodes. Our proposed approach is judged considering a number of real-life and synthetic datasets, where we consider ten state-of-the-art competitors. The experimental analysis shows superiority of the proposed approach over its competitors based on both accuracy and quality evaluations. Moreover, the detailed complexity analysis of the proposed approach shows linearly growth of the time complexity in terms of number of nodes and average degree, which is less than the same of majority overlapping community detection approaches.