http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/160 2026-03-10T15:16:17Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/48917 標題: 高覆蓋率中文關連樣式探勘以加速及完備知識圖譜之建立; Chinese Relation Patterns Mining with High Coverage for Knowledge Base Acceleration and Completion 作者: Sheng-Lun Wei; 魏聖倫 摘要: 近年來，隨著網際網路的迅速發展，人們可以透過不同的管道取得大量的資訊，例如：網路新聞、社群網路、部落格、論壇等。人們每天在網路上製造大量的資訊，其中有些資訊經過蒐集、整理、歸納後是值得被人們儲存並再次利用的。知識庫即是常用來儲存這類有用資訊的方式之一，人們多半使用結構化的方式來儲存，以利往後能夠更加方便的使用這些知識。 然而，由於多數知識庫皆由人類編輯彙整，在這資訊爆炸的時代，資訊產生的量遠高於志願編輯者所能負擔，使得從事件發生到被新增到知識庫中會有一定程度的時間間隔。因此，如何有效的加速知識庫的建立將會是個重大的課題。關連樣式是人們常用來加速知識庫建立的方式，但除了英文之外，很少有其他語言的關連樣式資源可以讓人們使用。 本研究提出一套建立高覆蓋率中文關連樣式庫的方式，以加速知識庫的建立以及知識庫的應用。本研究以DBpedia的實體特性作為依據，針對每個實體特性進行探勘，找出其對應的中文關連樣式。我們將詳細的說明每個步驟的實作細節，包含文本的前處理、實體範例擷取、以及關連樣式萃取共三個部分。此外，我們也會討論過程中可能出現的問題，以及這些問題的影響與解決方式。最後，本研究使用人工標記者去衡量中文關連樣式的效能，並討論不同因素對於關連樣式品質的影響。 以往人們可以藉由應用英文關連樣式庫做相關的研究，其他語言因沒有較完整地關連樣式資源不得其門而入。如今，可藉由本研究產生之高覆蓋率中文關連樣式庫進行相同領域的研究，讓知識庫相關的研究能夠不只在英文領域發展，也同樣能在中文領域開啟一片天。此外，雖然本研究提出的方式主要是針對中文關連樣式的建立，但我們認為其他和中文有類似特性的語言，例如：日文、韓文，皆可嘗試使用本研究提出的方法來建立該語言專屬的關連樣式庫。; With the rapid development of the Internet in recent years, people can get infor-mation from it through different sources such as online news, social network, and fo-rums. A lot of information is created by people every day and some of them can be col-lected, comprehended, and turned into knowledge by human beings. Knowledge base is a way that people store those information with structural format. However, it’s hard to keep knowledge base up-to-date because of the wide gap between limited editors and numerous information of entities. Knowledge base acceleration is a critical issue which focus on accelerating the construction of knowledge base. In addition, relation patterns are useful for knowledge base acceleration. However, there are no resources available in languages beyond English. In this study, we present a workflow for building relation pattern extraction system with high coverage for knowledge base acceleration and knowledge base completion. Our properties is based on the properties in DBpedia knowledge base. We will discuss many details of our method including corpus pre-processing, instance retrieval, and pat-tern extraction. Finally, we evaluate our relation patterns by human annotators and dis-cuss features that may affect the performance of the relation patterns. With Chinese relation patterns, many related work can be utilized in Chinese by transferring from English environment to Chinese environment. Other languages may also use our method to build their own relation pattern resources. 2016-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/26680 標題: 高效率主動式惡意軟體蒐集系統; An Effective Proactive Malware Collector 作者: Yuan-Tao Li; 李遠濤 摘要: 在網路蓬勃發展的今日，病毒傳播愈來愈快速，大量的新興與變種病毒不斷的產生，而且技術越來越精良：使用Rootkit方式執行的病毒，偽裝網路活動、註冊機碼、處理程序等所有可能警示使用者系統中潛伏著惡意程式的項目，隱藏在系統之中，一般人不易察覺；MSN病毒利用社群關係，降低使用者警戒心，達成以等比級數快速散播的驚人速度。因此，病毒研究員急需快速與大量的取得各種的病毒樣本，尤其是正在網路上散播的新型病毒，來進行分析，才能應付與日遽增病毒的威脅。 本篇論文提出了Proactive Malware Collects Tool，一個可以主動連接遭受感染的網站，自動擷取出受感染的樣本的工具。簡而言之，我們取得受感染網站的列表，並在模擬的作業系統環境下一一瀏覽這些網站，擷取出瀏覽該網站後新增、異動的檔案，再進行篩選，找出可能遭受感染的檔案，提供後續分析使用。 我們的工具利用比對虛擬機器底層檔案活動的差異，以未修改Windows環境的方式來偵測病毒活動產生的檔案，不易被病毒發現。此外，我們的工具從取得連結、瀏覽、篩選皆是自動化的。因此，Proactive Malware Collects Tool是一個自動化收集大量病毒的的理想工具。; Internet services are increasingly becoming an essential part of our everyday life. But the viruses spread more and more fast. Large numbers of new risen and new sophisticated viruses are constantly expanding, and their techniques are more and more compact. In the form of Trojan for example that aims to perform its tasks with user consent, and usually is disguised as a legitimate program – apparently it greatly compromises the integrity of the system. Users infected with Trojans cannot be aware of having infected. Another MSN worms use the social relationship to reduce the alert of users and spread at a amazing speed of doubling the number each square. Therefore, malware researcher urgently needs all kinds of malware samples for investigating, especially the new kinds of worms in the Internet. The better and more we know about what malware is currently spreading in the wild, the better can our defenses are. In this thesis, we describe a Proactive Malware Collector, a tool that connects the compromised websites, and automates to get the infected samples. In brief, we get the list of the compromised websites, and browse each site in an unmodified Windows environment, which leads to excellent emulation accuracy. We capture the created and modified files after browsing the sites and filter those files that could be infected for further in-depth analysis. To this end, our tool uses the technique that is comparing the difference of virtual hardware file activity for obtaining the infected samples. It is invisible to malware. Furthermore, our tool automates to get links, browse, and filter. These factors make The Proactive Malware Collector an ideal tool for automatically collecting the large numbers of malware. 2008-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/49587 標題: 高效且穩定收斂的平行非同步隨機對偶座標梯度下降演算法; Parallel Asynchronous Stochastic Dual Coordinate Descent Algorithms for High Efficiency and Stable Convergence 作者: Yung-Chen Chen; 陳勇辰 摘要: 平行非同步隨機對偶座標梯度下降演算法(PASSCoDe)是一種用於在單台多核心且共用記憶體的機器上訓練線性模型的演算法。 當執行緒的數目小於8時，PASSCoDe在稀疏資料集(非零項所佔的比例很低的資料集)上有著十分良好的加速。 但是因為記憶體的衝突及延遲參數存取的因素，PASSCoDe時常無法收斂到和循序隨機對偶座標梯度下降演算法相同的準確度或是無法收斂。 在此篇論文中，我們提出了兩個演算法 - 適應性混合算法及惰性同步算法以克服平行算法無法收斂的議題。實驗結果指出，我們所提出的兩個演算法在所有資料集中皆可以收斂至和循序隨機對偶座標梯度下降演算法相同的準確度除了一個極小的資料集例外。相比之下PASSCoDe收斂至一個較差的準確度或是不收斂。我們所提出的演算法在收斂的穩定度、執行時間和可擴展性上超越了PASSCoDe的一個改進過後的版本PASSCoDe-Fix.; Parallel asynchronous stochastic dual coordinate descent algorithm (PASSCoDe) is an efficient method to train linear models in multi-core shared-memory systems. PASSCoDe enjoys a good speedup when the number of threads is less than 8 on sparse datasets, i.e., the percentage of nonzero elements in the training data is relatively small. However, due to the memory conflict and delayed parameter access problem in parallel execution, it often diverges or does not converge to the best accuracy as a serial dual coordinate descent algorithm does. In this paper, we proposed two algorithms - Adaptive Hybrid algorithm and Lazy-Sync algorithm, to overcome the convergence issues in parallel execution. Experiment results indicate that both algorithms converge to the same high accuracy as a sequential program does on all datasets we tested, except on one extremely small dataset. On the other hand, PASSCoDe sometimes converges to a less accurate value or does not converge at all on some datasets. Our methods also outperform PASSCoDe-Fix, an improved version of PASSCoDe, in stable convergence, execution speed, and scalability. 2020-01-01T00:00:00Z http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/50022 標題: 類圓頂之超廣視野頭戴式顯示器系統; Dome+: A Dome-Like Ultra Wide Field-of-View Head-Mounted Display System 作者: Pei-Hsuan Tsai; 蔡沛軒 摘要: 本論文提出一個新型圓頂狀頭戴式顯示器系統 — Dome+，使用了9個面板拼接而成之顯示器，提供215度超廣水平視野來產生沉浸式的虛擬實境體驗。藉由計算不同使用者之頂點距離與視覺調節力，讓近視與沒近視之使用者皆不需戴眼鏡即可使用Dome+系統，同時也消除一般使用於頭戴式顯示器中透鏡散射所產生的色差問題。此研究也提出幾種原型應用，及利用統計方法威爾克森符號等級檢定，評估雙眼立體視覺顯示器與非立體視覺的大視野彎曲顯示器在虛擬實境體驗中的沉浸程度，得到了即使沉浸傾向低的使用者也能在Dome+系統之超大角度視野下產生自然三維效果(Natural 3D effect)，感受到立體感且更加身歷其境。本篇論文提出的超大角度視野頭戴式顯示器系統，在虛擬實境中提供了一個新方式來達到更加的沉浸式經驗。; This thesis presents a novel dome-shaped head-mounted display (HMD) system, called Dome+, which uses 9 tiled panels for display to offer an ultra wide field-of-view (FoV) of 215 degrees for an immersive virtual reality (VR) experience. By calculating the vertex distance and the accommodation power of users, those who have myopic eyes or normal eyes can use Dome+ with their bare eyes; therefore we can eliminate the chromatic aberration artifacts which happened in current HMDs caused by lens dispersion. This work also introduce several prototype applications and use the non-parametric statistical analysis of Wilcoxon Signed Rank Test to analyse how immersive between the stereo display and the non-stereo, wide FoV curved display VR experiences. The results show that Dome+ offers a natural 3D effect even for those who have the weak tendencies to become involved in and so can feel stereoscopic effect. The proposed ultra wide FoV HMD system provides a new solution to achieve a more immersive VR experience. 2016-01-01T00:00:00Z