Local community detection using greedy algorithm with probability

Abstract

With the arrival of the era of big data, the scale of network has grown at an incredible rate, which has brought challenges to community discovery. Local community discovery is a kind of community discovery that does not need to know global information about network. A quantity of local community discovery algorithms have been put forward by researchers. Traditional local community discovery generally needs to define local community modularity Q, and greedily add nodes to the community when ΔQ > 0, which is easy to fall into the local optimal solution. Inspired by the ideal of simulated annealing, greedy algorithm with probability LCDGAP is proposed to detect local community in this paper, which can be applied to all algorithms that perform greedy addition. We permit that the node can be aggregated into the community with a certain probability when ΔQ < 0. At the same time, we guarantee that this probability will be getting smaller with the increase of program running time, ensuring the program’s convergence and stability. Experimental result proves that LCDGAP performs effectively not only in real-world dataset but also computer-generated dataset.