由於電影工業的蓬勃發展，以及數位化視訊分析與儲存技術的進步，使用者可藉由DVD所提供的故事分段索引來快速瀏覽及搜尋影片。因此一套能對電影作故事分段的工具是不可或缺的。 / 本論文的研究目的是針對人際關係類型的電影做故事單元的分段。我們提出以角色的社會網絡為基礎的方法，作電影故事單元的分段。此方法包括四個步驟。首先對電影作場景變換偵測。接著，我們利用人臉辨識技術擷取出每一個場景出現的角色。第三，我們考慮角色重要性對分段的影響，利用社會網絡分析中計算角色的網絡中心性，來衡量角色的重要性。最後，以角色為特徵值，比對場景之間的角色來計算相似度，並且利用循序性叢集分析，來達到電影故事單元的分段。我們的實驗針對四部人際關係類型的電影，以切成故事單元來評估分段的效果，實驗顯示以角色的社會網路為基礎的方法，準確率介於0.63到0.94之間。 / With the progress of entertainment industry, and the advances of digital video analysis and storage technologies, users can utilize the indexes of the DVD chapters for quick access, retrieval and browsing of movie content. Therefore, development of automatic movie content analysis is important. In this thesis, we focus on the romance and relationship movies, which contain the narrative of the relation between peoples. We propose a novel method for film narrative exploration based on social network analysis of characters. There are four steps. First, we perform movie scene change detection to segment a movie into scenes. In the second step, we extract the characters as the feature model of scenes by utilizing the face recognition system. Then, we measure the weight value of the characters by the centrality of social network analysis. Finally, we calculate the cosine similarity between scenes, and segment a movie into story units by using sequential clustering algorithm. We conduct experiments on four romance and relationship movies. Experimental result show that our proposed story unit segmentation method based on social network analysis of characters achieves from 63% to 94% accuracy.