社群網路為由於其廣泛的實用性而引起了研究者的關注。開發了幾種技術用於從社群網路之使用者的規律中探勘有用的知識。意見領袖發現是一項具有重大商業和政治價值的重要任務。通過找尋意見領袖，公司或政府可以分別利用其強大之影響力進行銷售活動或指導公眾輿論。此外，檢測有影響力的評論能夠理解輿論形成的來源和趨勢。然而，之前的研究主要集中在尋找靜態社群網路中的意見領袖，較少考慮社群網路隨時間演進的影響。在實際應用中，社群網路通常隨著時間的推移而演變，在動態社交網絡中尋找意見領袖的研究工作很少且具備難度。在本文中，提出了一種新的尋找意見領袖方法: DOLM，以有效地從動態社群網路中識別意見領袖。我們首先構建動態社群網路，然後檢測社區結構以解決資訊重疊問題。然後，DOLM開發基於分群的領導力分析，以找出動態社群網路中的意見領袖。實驗研究表明，該方法能夠有效捕捉動態社群網路的特徵，解決資訊重疊問題。最後我們還在幾個真實資料集上應用DOLM，以檢視意見領袖探勘的效率和擴展性。;Social network analysis has attracted researchers’ attention due to its widespread practicability. Several techniques are developed for extracting useful knowledge from users’ regularities. Opinion leader discovery is one essential task which has great commercial and political values. By identifying the opinion leaders, companies or governments could manipulate the selling or guiding public opinion, respectively. Additionally, detecting the influential comments is able to understand the source and trend of public opinion formation. However, prior studies mainly focus on finding opinion leader in a static social network. Actually, in real applications, social networks are usually evolved with time; few research efforts have been elaborated on finding opinion leaders in dynamic social network. In this study, a novel algorithm, DOLM, is proposed to efficiently find the opinion leaders from a dynamic social network. We utilize a network emerging method to construct a dynamic social network, and then detect the community structure to tackle the information overlapping problem. Then, DOLM develops a clustering-based leadership analysis to find out the opinion leader in a dynamic social network. The experimental study shows that the proposed algorithm could effectively capture the characteristic of a dynamic social network and solve the information overlapping problem. We also apply DOLM on several real datasets to show the efficiency and scalability for opinion leader discovery.
