在現代研究中，網路模型已經被應用在許多不同的領域中。例如社會學(Social science)、代謝網路(Metabolic network)、生物學，網頁圖(Webpage graph)和網際網路等等的網路。在研究網路模型上，有很多的網路特性被用來量測網路模型，例如群聚性，節點度分佈，平均網路路徑長度和s-度量標準。依據網路特性量測的結果，網路模型除了被描述成隨機網路(Random netowrks)以外，而且也可以描述成小世界網路(Small-world networks)或者無尺度網路(Scale-free networks)。因此，通信網路的分析和模型化，可以利用量測網路特性的方式作為分析或是建立網路模型的依據，以期結果能接近實際網路。為了要了解目前現行的網路的特性以及將來設計網路通訊協定時的依據，一個精確的通訊網路模型是一個很重要的課題。 在這篇論文中我們量測通信網路，並且基於量測的結果提出建立網路模型的方法。在論文中主要是探討的議題是如何建立通信網路模型以及利用已知的各種網路特性來評估分析網路。我們提出二個建立人際通信網路模型的方法。這些方法是利用社群緊密關係(Socail affinity)的概念來實踐在實際網路中的人際的緊密關係特性。 通常網路特性都只是統計的結果，因此網路特性並沒有辦法分辨網路拓撲之間的不同。在無尺度網路中，冪次律分佈(Power-law distribution)只能表示網路中的節點分支度分佈的特性。因此，可能有無尺度網路具有相同的網路冪次律分佈特性，而事實上這些網路的網路拓撲是不同的。例如，假設在一個網路中有二個分離的連結(edge)。這二個分離的連結可以重新連接成新的二條連結並且保持原來的節點分支度。這樣就會有二個有不同網路拓撲的網路，但是具有相同的節點分支度。我們提出連結交換演算法(Edge switching algorithm)來研究在具有相同網路特型的網路之間，這些網路的差異性。依據我們的模擬結果顯示，我們可以在無尺度網路中可以調整網路叢集度以及s-度量標準的高低，但調整後的網路仍保有相同的無尺度網路的特性。 Network models are commonly used in many branches of science, such as social networks in sociology, web page graph and Internet in computer science, etc. Many network properties are proposed to studying network models, such as clustering, node degree distribution, network path length and s-metric. From network properties, network models are not only described as random networks but also described as small-world networks or scale-free networks. The analysis and modeling of communication networks therefore rely on those measurements of networks to verify that the generated network models are similar to real data. Accurate modeling of communication networks is an important issue in both understanding the current networks and designing future communication protocols. This thesis measures communication network data and proposes network models based on those measurements. Issues covered in this thesis include the modeling of communication networks and the evaluation and analysis of networks based on some well-known network properties. Two models are proposed to modeling human communication networks. The proposed network models use the concept of social affinity to capture the closeness property presented in real world networks. Usually, network properties are based on aggregate statistics and are not good indicators to differentiate different topologies. In scale-free networks, the power law distribution only related to graph connectivity. Therefore, the scale-free networks which have different topologies may have equal power-law degree distributions. For example, assume that, there are two separate edges in a network. The two edges can be rewired and preserved node degree. And then, there exist two networks which have different topologies and node degree preserving. Edge switching algorithms are proposed to studying differences between networks that have same network properties. The simulation results showed that the clustering coefficient and s-metric of scale-free networks can be tuned while preserving scale-free property.
