| 摘要: | 中文摘要
偵測真實網路中的community結構是重要的研究主題，在許多方面皆有所應用，譬如:生物學、社會學、全球資訊網以及行動通信網路等。有許多不同的community偵測方法被提出。其中，我們發現Palla, G., et al提出的k-clique community相當重要。不過，他們的程式CFinder執行時間比起其它的偵測方法都要來得長。我們提出偵測k-clique community的平行演算法Parallel-Community，並實作在多核心電腦上，藉以縮短偵測k-clique community的時間。不過，在我們實作的時候，發現我們的作法和Gregori, E., et al的COS演算法，在某些部分作法相似。我們用五張真實網路圖，來測試比較Parallel-Community、COS和CFinder。使用單處理器來執行時，Parallel-Community與CFinder相比，最快可以到130倍，最慢至少可以達到54倍。使用多處理器來執行時，Parallel-Community與COS相比，最快可以到61倍，最慢至少可以達到24倍。;Abstract.
Detecting communities is an important research topic. It can be applied to biology, sociology, World Wide Web, mobile communication networks, etc. Many different detection methods have been published. Among them, k-clique community by Palla, G., et al is an important one. However, their program Cfinder’s execution time compared with other detection methods is time consuming. In order to reduce the running time of detecting communities, we propose a parallel algorithm of k-clique community "Parallel-Community" and implement it on a multi-core parallel computer. When we implement the algorithm, we find out some parts of Parallel-Community are similar to Gregori, E., et al’s COS. We experiment on five real networks to test Parallel-Community, COS, and CFinder. When executing with single processor, Parallel-Community achieve 54 to 130 times faster than CFinder. When executing with multiple processors, Parallel-Community achieve 24 to 61 times faster than COS. |
