近年來，隨著Web 2.0 概念的蓬勃發展，社群網路儼然成為現代人最主要的接收資訊的管道， 再加上電子商務的普及化，社群網路中的意見領袖探勘與影響力最大化問題，吸引了許多專家學 者的目光。然而現今的研究方法中，因社群網路的複雜性，造成了演算法的效能不佳，尤其當我 們所要面對的是大量的社群資料時。 本計畫將著墨在有效率的意見領袖探勘與影響力最大化問題，設計演算法與研究應用領域， 內容概述如下： 1. 社群網路群集偵測方法：我們設計了一個能快速，且不需輸入任何參數的群集偵測演算法，且 能正確地找出社群網路中的群集架構。 2. 有效率的意見領袖探勘演算法：利用所提出的群集方法，我們設計了一個結合領袖特質分析與 網路結構分析的演算法，能有效率的探勘社群網路中的意見領袖。 有效率的影響力最大化演算法：利用所提出的群集方法，我們設計了一個利用逆向可達集合與網路結 構分析的演算法，能快速的找出影響力最大化的種子集。 ;Recently, owing to the concept of Web 2.0, many social websites have become part of our daily life. Hence, the analysis of social network attracts many researchers’ attentions. The opinion leader discovery and influence maximization are two hot topics due to their applicability. However, due to the complexity of graph processing, it is still a challenge when handling a huge network. The efficiency is still a fatal problem for many prior studies. In this project, we will focus on three topics, (1) Community detection: We propose an efficient and parameter-free clustering algorithm to efficiently discover the community structure of a social network. (2) Opinion leader mining: We combine the leadership-analysis-based and network-structure-based methods and develop a novel algorithm for efficiently mining opinion leader in a social network. (3) Influence maximization: Based on the proposed community detection method, we utilize the community information and reversed reachable set to efficiently find the seed set that could maximize the influence spread in a social network. Furthermore, the proposed algorithms are applied on real dataset to show the practicability and efficiency.
