基于连接密度的网络社团发现方法分析与实现-analysis and implementation of network community discovery method based on connection density.docxVIP

摘要摘要摘要伴随着互联网及相关信息技术的迅猛发展，人类社会已经迈入了“网络时 代”。复杂网络（Complex Network）系统随处可见。复杂网络的一个重要属性是 其内在的社团结构。挖掘网络的社团结构，有助于揭示网络拓扑结构，分析其功 能和隐含模式，预测复杂网络的行为，在社会网络、生物信息、搜索引擎、信息 推荐、精准营销等领域都有较大的应用价值。本文首先深入研究了网络社团的定义、结构及当前主流的网络社团发现方法， 分析其优缺点。基于此提出了一种基于节点连接密度的局部社团质量评价标准， 并据此提出一个快速在线多分辨率重叠社团检测算法。该算法无需网络全局信息 和用户干预，能够从网络中的任一节点开始，自动检测包含起始节点的局部社团。 其次，本文将基于连接密度的网络聚类方法和基于最小生成树的网络聚类方法相 结合，提出了一种基于网络骨架树的密度网络社团发现方法。该算法能准确的识 别出各种规模的社团以及中心点和离群点,并能定位最优密度阈值。最后,在真实网络和生成网络上对算法进行了详细的对比实验。结果表明：本 文提出的网络社团发现方法准确率高，运行速度快，能够处理各种规模和形状的 网络。关键词 社团发现 局部社团 密度聚类 网络骨架树AbstractAbstractAbstractWith the development of internet and related information techniques, the human society is entering the age of networks. Complex networks are everywhere around us. One of the most important properties of the complex network is the internal communities of it. Mining the community structure of the complex network can help us reveal the topology structure of networks, analyze their function and hidden pattern, and predict the action of complex network, which are valuable in search engine, recommender system, Precision marketing and so on.In this thesis, we investigate the definition, structure and the mainstream method of network community detection at first. Based on this, a local community quality criterion is proposed as well as a fast online multi-resolution overlapping community detection algorithm. The algorithm proposed does not need any global information of the network and user intervention, can identify local community from a given start vertex. Then, another new community identifying method based on the skeleton tree of the network is proposed by combination of density based clustering method and minimum spanning tree based clustering method. The proposed method can identify any size of community as well as hubs and outliers, and locate the optimal density threshold.At last, detail material experiment is made on some real world and manmade networks. Results show that methods proposed in this thesis are accurate and fast, can p