基于LeaderRank的多标签传播重叠社区发现算法*

针对多标签传播重叠社区发现算法COPRA存在的的随机性强、鲁棒性差等问题，提出一种基于多标签传播思想的重叠社区发现算法。该算法通过LeaderRank算法来量化网络中节点的重要性，再根据量化值大小对节点进行团扩展，得到可重叠的最具重要性的粗糙团，分别对粗糙团和非粗糙团中的节点进行标签初始化，再通过合理的标签迭代顺序和改进的标签删选策略进行标签更新，直到达到标签传播的终止条件结束迭代过程。在人工网络图和真实数据集上进行实验，结果表明所提算法不仅有效地增强了社区发现结果的稳定性，同时提高了准确率。

Multi-label propagation algorithm for overlapping community detection based on LeaderRank

In order to avoid the strong randomness and poor robustness of COPRA(Community Overlap PRopagation Algorithm), this paper proposed a multi-label propagation algorithm for overlapping community detection. It used LeaderRank algorithm to quantify the importance of nodes in social network, then extended these nodes to maximal cliques according to LeaderRank scores, and initialized labels and belonging coefficients for all nodes, and then updated labels of all nodes by a reasonable iteration sequence and an improved label updating strategy until reaching the end condition. Experiments on LFR benchmark networks and real networks show that the proposed algorithm not only improves the stability effectively, but also increases the accuracy for detecting overlapping communities.