一种基于层次约简的多层网络社区发现算法

如何在多层网络中发现社区是一项巨大挑战。目前有些算法将多层网络表示成三阶张量，然后使用非负张量分解进行社区发现。但在多层网络的每层网络中存在很多社区之间的连接或每层网络都很稀疏的情况下，非负张量分解算法的准确率较差。为了解决这一问题，本文提出一种改进算法。先将原始多层网络进行层次约简，减少多层网络的层数，使其社区结构更加凸显，然后再使用非负张量分解算法进行社区发现。在人工数据集与真实数据集上的实验表明，本文所提出的框架在准确率上有明显的优势。

A Community Detection Algorithm of Multiplex Networks with Layer Reduction

How to detect community in a multiplex network is a knotty problem. Currently some algorithms represent the multiplex network as a three-way tensor and use non-negative tensor factorization to capture the community structure. However, if there are many edges between communities or when the multiplex network is sparse, the non-negative tensor factorization algorithm won’t work well. To this end, this paper introduced an improved algorithm. The algorithm first merges the layers which have strong correlation to reduce the number of layers of multiplex network for the sake of highlighting the community structure. And then the algorithm uses non-negative tensor factorization to detect community. This paper validates the approach on both synthetic benchmarks and real multiplex networks, and the result shows that the algorithm performs better than the old approach.