多特征融合的标签传播算法

Label Propagation Algorithm Based on Multi-Feature

论文在分析常用社区发现算法的优缺点时,指出了标签传播算法(LPA)具有时间复杂度低、不需要预先设置社区个数、计算过程简单,在处理大型复杂网络时,具有较高的效率的特点。但该算法在标签传播的过程中,未考虑到相邻节点在网络结构以及内容中的相似性。因此论文从节点相似度角度出发,提出了多特征融合的标签传播算法。该算法首先利用SimRank算法计算网络中节点的结构相似度,同时使用主体模型得到节点内容的主题分布,并计算不同节点主题分布的相似度,最终融合两种相似度,为邻接节点传播来的标签,赋予相应的权重,以此来改进传播策略。实验比较,该算法较优于传统的标签传播算法。

When the advantages and disadvantages of common community discovery algorithms are analyzed,this paper points out that the label propagation algorithm(LPA)has the advantages of low time complexity,it doesn’t need to set the number of communities in advance,simple calculation process,high efficiency in dealing with large complex networks.However,in the process of label propagation,this algorithm does not consider the similarity of adjacent nodes in the network structure and the content.Therefore,this paper proposes a multi-feature fusion label propagation algorithm from the perspective of node similarity.The algorithm firstly uses SimRank algorithm to calculate the structural similarity of nodes in the network.At the same time,the main model is used to obtain the topic content distribution of the nodes,and the similarity of the topic distribution of different nodes is calculated.Finally,the two similarities are merged,and the labels propagated by neighboring nodes are given corresponding weights to improve the broadcast strategy.Through experimental comparison,this algorithm is superior to the traditional label propagation algorithm.