现实世界的大部分网络，比如社交网、因特网、交通网、神经系统、大脑等，都可以将其视作为复杂网络，其由许多节点和边组成。模块化社区是复杂网络的基本性质。社区主要指，内部彼此之间连接稠密，但外部（社区与社区之间）连接稀疏的簇。复杂网络的社区挖掘主要是基于网络的拓扑结构去识别网络模块（社区），可分为非重叠社区挖掘（一个节点只能从属于一个社区）和重叠社区挖掘（节点可以从属于一个或者多个社区）。许多现实世界的网络，比如社交网络、因特网等都可认为是重叠的。有研究表明，脑网络可能也具有重叠社区属性，因此基于重叠社区挖掘算法的脑网络分析也许能为认识脑、理解脑的机制提供新思路。 本研究主要基于功能性磁共振成像（functional Magnetic Resonance Imaging, fMRI），利用重叠社区挖掘算法进行脑网络分析，主要工作和研究结果如下： 一、基于模拟数据的重叠社区挖掘算法评价。为了分析重叠社区挖掘算法的具体性能和适用条件，首先基于LFR（Lancichinetti Fortunato Radicchi）基准网络以及真实社交网络进行重叠社区挖掘算法评价，并且探讨了SLPA（Speaker-Listener Based Information Propagation Process Algorithm）算法后处理过程中阈值的选取问题。结果表明：SLPA算法后处理阈值r在0.2至0.25之间，所得到的社区结构最好。随着重叠节点所属社区数、网络混合系数、重叠节点所属社区数的增加，SLPA算法和COPRA（Community Overlap PRopagation Algorithm）算法的NMI（Normalized Mutual Information）值和EQ（Extended Modularity Q）值均有所下降。总体上，SLPA算法性能优于COPRA算法，普适性更强。 二、基于SLPA算法在fMRI数据集上的重叠脑网络分析。主要基于两个静息态fMRI公开数据集展开研究，首先，探讨了基于静息态fMRI数据SLPA算法的参数选取问题。其次，对二值化和带权功能连接矩阵的社区挖掘结果差异以及两个数据集之间的结果差异进行了分析。最后，进行了重叠脑区和重叠脑网络的分析工作。结果表明：当阈值r为0.2至0.25时，SLAP算法挖掘效果较好；基于带权功能连接矩阵的社区挖掘效果比二值化功能连接矩阵的效果更好无，并且两个数据集之间的重叠社区挖掘结果差异不显著，验证了SLPA在两个数据集实际应用中的稳定性。高频次重叠脑区主要为额叶、丘脑、海马旁回、颞中回、楔前叶、枕叶、脑岛以及梭状回等，并且高频次重叠节点具有更高的度中心、节点效率、聚类系数、节点局部效率等。重叠脑网络分析结果表明，部分社区与视觉网络、感觉运动网络、腹侧注意网络相似。多个网络之间存在重叠。 综上，本研究不但验证了SLPA算法在脑网络研究中的稳定性和可靠性，发现带权功能连接矩阵确实能够更好地进行脑网络构建，而且针对重叠脑区与重叠脑网络的特性和具体认知功能进行了详细分析，为未来大脑网络的分析工作提供了新角度。

Most of the real world networks, such as social networks, the Internet, transportation networks, nervous systems and human brain can all be regarded as complex networks, which consists of many nodes and edges. The modular community is a fundamental property of complex networks. The community mainly refers to the clusters that dense connections between the internals, but the external (between the community and the community) are sparsely connected. The community detection of complex networks is mainly based on the network topology to identify network modules (communities), which can be divided into non-overlapping community detection (a node can only be subordinate to a community) and overlapping community detection (nodes can be subordinate to one or more communities). Many real-world networks, such as social networks, the Internet, etc., can be considered overlapping. Studies have shown that brain networks may also have overlapping community attributes, so brain network analysis based on overlapping community detection algorithms may provide new ideas for understanding the brain and understanding the brain's mechanism. This study is based on functional magnetic resonance imaging (fMRI), using overlapping community detection algorithms for brain network analysis. The main work and research results are as follows: First, evaluation of overlapping community detection algorithms based on simulated data. In order to analyze the specific performance and applicable conditions of overlapping community detection algorithms, firstly, based on LFR (Lancichinetti Fortunato Radicchi) benchmark network and real social network, the overlapping community detection is carried out, and the selection of threshold r in post-processing process of SLPA (Speaker-Listener Based Information Propagation Process Algorithm) algorithm is discussed. The results show that when the threshold r is between 0.2 and 0.25, and the detected community structure is the best. With the number of communities to which the overlapping nodes belong, the network mixing coefficient, and the number of communities to which the overlapping nodes belong, the NMI (Normalized Mutual Information) value and the EQ (Extended Modularity Q) value of the SLPA algorithm and the COPRA (Community Overlap PRopagation Algorithm) algorithm are both declining. Overall, SLPA algorithm performance is better than COPRA algorithm, and the universality is stronger. Second, the overlapping brain network analysis based on SLPA algorithm on fMRI dataset. Based on the research of two resting state fMRI open datasets, firstly, the selection of threshold r in post-processing of SLPA based on resting state fMRI data is discussed. Secondly, the difference of community detection results between the binarized functional connection matrix and the weighted functional connection matrix and the difference of community detection results between the two data sets are analyzed. Finally, the analysis of overlapping brain regions and overlapping brain networks was researched. The results show that when the threshold r is 0.2 to 0.25, the detection effect is better. And the community detection performance based on the weighted function connection matrix is better than the binarized function connection matrix. The overlap community detection results between the two data sets are not significantly different. The high-frequency overlapping brain regions were mainly frontal lobe, thalamus, hippocampus, sacral gyrus, anterior wedge, occipital lobe, insula and fusiform gyrus, which have higher degrees centrality, node efficiency, clustering coefficient, node local efficiency, etc. Some communities are similar to visual network, sensorimotor network and ventral attention network. There are overlaps region between multiple networks. In summary, this study not only verifies the stability and reliability of the SLPA algorithm in brain network research, but also finds that the weighted functional connection matrix can better construct the brain network, and the characteristics and the cognitive function of overlapping brain regions and overlapping brain networks has been analyzed in detail, which provides a new perspective for the future analysis of brain networks.