动态网络拓扑结构变化的多角度度量-计算机应用技术专业论文.docxVIP

摘要 动态网络的拓扑结构在整个研究阶段内必定在不断演化改变，对于很多应用而 言，无论是准确描述动态网络整体变化趋势还是更精确的计算网络间变化程度都 是很有意义的工作。 目前，已有许多工作致力于比较图之间的相似性或距离，然而很多经典算法存 在两个特点，一方面它们大多针对无标记图，导致算法时间复杂度太高；另一方 面它们只考虑图之间顶点和边的差异量，对图距离的定义过于狭隘。本文针对顶 点唯一标记的复杂网络，提出四种图距离计算方法，包括基于顶点相关性、基于 交互相关性、基于模块相关性和基于谱距离，分别从网络拓扑指标和特征系统度 量动态网络的拓扑结构变化情况。同时，还明确给出动态网络持续性和波动性的 定义，用来评价网络整体变化程度。通过计算所有相邻时刻快照间图距离，既可 以了解网络整体变化趋势，又可以检测时序网络的异常点，这在包括复杂疾病恶 化预测或信息网络异常检测等领域有良好应用前景。 本文在人工合成网络数据和真实网络数据上对提出的四种方法进行了实验仿 真，实验结果表明不同方法确实能够有效描述网络不同方面变化特点。为进一步 研究动态网络的变化平滑性，本文对动态网络所有快照进行随机洗牌处理，通过 洗牌前后网络持续性和波动性值的比较，从中发现一些有趣的结论。 关键字：动态网络 图距离 拓扑指标 特征系统 Abstract The topology of a dynamic network is inherently undergoing evolution and change during the observation period. For many applications, it is of great value no matter to accurately describe the overall evolution trend or to precisely measure the changing level between networks. Currently many existing works concentrate on comparison of graph similarity or distance. However, many classical methods own two inappropriate characteristics. One thing is that they focus on unlabeled graphs, which results in high time complexity, and the other is that they only define the volume of different vertices or edges as the distance, which is not enough in some cases. Focusing on vertex uniquely labeled complex networks, this work proposes four graph distance computation methods which are vertex correlation based, interaction correlation based, module correlation based and spectral distance based. These methods measure the topological structure change of dynamic networks from network topology indicators and eigensystem respectively. In the mean time, we formally give the definitions of persistence and volatility in dynamic networks to evaluate the overall change degree. After measuring the graph distance between every two consecutive snapshots, we can not only learn about the overall change trend, but also detect the anomaly time points, which have an excellent application prospect in many fields, such as complex disease progre