鄭卜壬臺灣大學:資訊網路與多媒體研究所朱崇賓Chu, Chun-BinChun-BinChu2010-05-052018-07-052010-05-052018-07-052009U0001-1808200915282400http://ntur.lib.ntu.edu.tw//handle/246246/180742用事件流程圖的方式來理解新聞發展的來龍去脈,比直接去瀏覽整個相關新聞列表來的有效率且容易吸收.而要建立一個事件流程圖,我們必須先找出新聞事件之間的關係,才能將它們串接起來.篇論文提出了一個能夠自動偵測事件關係的方法.我們的方法是基於兩個事件的相似度來判斷它們是否具有前後關係.有別於單純的相似度運算,我們不只考慮到事件的主題,也考慮到事件中名稱實體所扮演的角色.在這篇論文中,主題是用來描述這個事件的輪廓,我們可以用主題來了解這個事件是在屬於哪一類型的新聞.而角色是代表一個命名實體在新聞事件中的行為,處境..等概念,即 代表一個實體在新聞事件中所扮演的角色.篇論文會介紹我們如何運用主題及角色的概念來偵測新聞事件的前後關係,以及做一些實驗來證明這個方法的可行性.Reading the evolution graph of news events is much easier than browsing the news events in a list. To construct the evolution graph, we have to detect the dependent relationship between news events.n this paper we present a method to automatically detect the evolution of news events. Our method is based on the similarity between news events, our method considers not only the characteristic of news events, but also considers the roles played by entity names in news events. The characteristic of news events could be defined as the topic of these news events, we can roughly understand what happened about these news events by topic information. The role means the concept behind a name entity which describe the name entities’ behavior, situation…etc. The same entity name might play different roles in different news events. We will introduce how we use the role and topic information to detect the relationship between news events.摘要 iibstract iiihapter 1 Introduction 1.1 Motivations 1.2 Previous Work 3.3 Problem Definition 4.4 Basic Idea 5.5 Challenges 6.6 Thesis Organization 7hapter 2 Relate Work 8.1 Topic Detection and Tracking (TDT) 8.2 Event Threading 10hapter 3 Model the News Article 12.1 Analysis of News Articles 12.2 Modeling the News Article 13.3 The Role 14.4 The Topic 21.5 Review of the Model 22hapter 4 Relationship Detection 24.1 Operations 24.2 Similarity between Documents 25.3 Role Similarity and Topic Similarity 29.4 Discussion 30hapter 5 Experiments 31.1 Overview 31.2 Dataset 32.3 Evaluation 33.4 Baseline 34.5 Experiment 1. Different Weight 35.6 Experiment 2: Window Size 43.7 Experiment 3: Different Feature Word 45.8 Experiment 4: Different Feature Size 46hapter 6 Conclusion and Future Work 48.1 Conclusion 48.2 Application 49.3 Future Work 50application/pdf1169273 bytesapplication/pdfen-US事件關係偵測EventDetectionRelationshipthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/180742/1/ntu-98-R96944038-1.pdf