Abstract: Nonnegative matrix factorization (NMF) has good ability in extracting inherent patterns and structures in high dimensional data and has been one of hot research topics in data mining. Nonnegative matrix factorization is a tool for unsupervised learning and has been widely applied in pattern recognition, text mining, image processing and bioinformatics. Recently, many researchers have paid attention to network-based data mining via nonnegative matrix factorization in order to detect cohesively connected community in networks. This paper summarizes community detection algorithms using nonnegative matrix factorization, including unsupervised methods and semi-supervised algorithms. Then, this paper compares and analyzes the performance of different algorithms by conducting experiments on artificial networks and real-world networks. Finally, this paper discusses challenges and further work on detecting communities in networks by using nonnegative matrix factorization.

Key words: data mining, nonnegative matrix factorization, community detection

摘要： 非负矩阵分解（nonnegative matrix factorization，NMF）在提取高维数据中隐含模式和结构方面具有良好性能，已成为数据挖掘领域的热点研究之一。NMF作为无监督学习的有效工具，在模式识别、文本处理、多媒体数据分析以及生物信息学等研究领域得到了广泛应用。目前，已有工作将NMF模型应用于网络数据挖掘，发现网络中隐含的社区结构。对基于NMF的社区发现方法进行了总结，包括无监督的社区发现方法和半监督的社区发现方法，通过在实际网络和人工网络进行实验，比较分析了不同算法的性能，进一步研究了当前基于NMF发现社区结构所面临的挑战，并对下一步研究方向进行了展望。

关键词: 数据挖掘, 非负矩阵分解, 社区发现