Figure 3 illustrates the process of community detection using algorithm NILP in the above example network when α = 2. In Figure 3(a), in the sample network,

each node is marked with a unique label, and the 2-degree neighborhood impact values are labeled beside the nodes. According to the ascending sort order of the impact values, the nodes update order is determined as 5 → 1 ALK targets → 4 → 2 → 3 → 6 → 7 → 8 → 9 → 10. Node 5 is the first one for label update, using formula (5) to decide the new label, and the result for adjacent neighborhood node 6 has the greatest influence on it, so we change the label of node 5 to the node number of its neighbor, in case 6. Next, we update all the nodes sequentially. Figure 3(b) is the result of the divided community which is updated at the end

of the first round of label propagation. After the first round of label update process completed, with the stable ratio of the current node being p1 = 0.3, we are supposed to update labels in accordance with the above order in the next round of node label update process. The algorithm continues to run until the stable ratio no longer rises. Figure 3(c) shows the final results of our algorithm on detecting communities on the sample network. Figure 3 The process of label propagation by using algorithm NILP to detect community structure on the sample network. Algorithm NILP is different from other label propagation based algorithms. First, NILP limits the scope of impact that nodes can exert on their neighbors to a variable α, and it differs from the attenuation degree setting in the label propagation process of LHLC, rendering it feasible for nonattenuation propagation in local areas

in real life. Such as a network of friends, only a limited number of people within the scope of the friends will be in the same circle of friends. When the information of insiders’ interest has been released, the information exchanges along the route of various relationships to attain the goal of information sharing, while outsiders are mostly not likely to disseminate such information because they are not interested in it. Secondly, NILP calculates Cilengitide the mean value of impact for each node in the scanning range of α-degree neighborhood and fully takes its α-degree neighborhood network structure into account, which improves the efficiency of the process of label propagation. Third, the mutual influence between nodes is an objective existence, independent of the label propagation, so the node neighborhood impact and the label iterative update process are separated. Due to the fact that label propagation proceeds with nodes affecting each other, the process of node update must be based on the value of average neighborhood impact. Finally, according to the size of neighborhood impact, NILP updates all the nodes in ascending order and makes the process of updating labels more definite instead of more randomized.