Research Specifications

Community detection is an appropriate approach for discovering and understanding the structure and hidden information in complex networks One of the most critical issues in the community detection problem is the low-time complexity of the algorithm while preserving the accuracy of the algorithm, which is important in large-scale networks such as social networks. Local community detection algorithms try to use local information and provide acceptable results in a reasonable time. This paper proposes a fast and accurate community detection algorithm based on local information for the community’s label assigning. In the proposed algorithm, local community detection is started from low degree nodes by label assigning in a multi-level diffusion way, called LSMD algorithm, with significant low time complexity. In the first phase of the LSMD algorithm, at first, the community’s label is assigned to the node with degree 1 and its direct neighbor and second level neighbors. In fact, we used this fact that people with a smaller number of neighbors are not likely to be connected to diverse communities. Next, a community label is respectively assigned for the nodes with degrees 2, 3, and so forth. In the second phase, initial communities are merged using a new simple and fast strategy as far as possible to form the final communities. Besides, there is no random nature in the algorithm, as well as the adjustable parameter. Therefore, the obtained results show meaningful stability for the LSMD. Experiments are performed on real-world and synthetic networks to evaluate the performance and accuracy of the proposed algorithm. The results show that the proposed algorithm significantly is more accurate than the other state-of-the-art algorithms. In addition, it substantially is faster than other local algorithms such as LPA, LCCD, NIBLPA, DA, RTLCD, Louvain, G-CN, Infomap, and ECES on large-scale networks.