隨著文字資訊的爆炸式增長，越來越多的訊息開始以電子文本的形式儲存及傳遞。但隨著文本內容資訊量不斷地增加，使用者也越來越難以快速地掌握文本全貌。因此本研究試圖透過主題模型（TopicModels）、標記式主題模型（Labeled Topic Models）演算法－在自然語言處理領域裡文本探勘的方法，識別出大規模文本中潛藏的主題訊 息之後，再利用圖像視覺化在資訊表達上的優勢和效率，透過各種視覺化圖案的呈現從不同的角度來探索文本，形成一種嶄新的大規模文本閱讀與分析方式。 本研究設計了兩階段實驗：第一階段任務導向性實驗、第二階段指定任務實驗，以及評估問卷來驗證本介面的易用性（ Ease-of-use ）和有用性（ Usefulness ）。並透過實驗問卷的分數結果驗證了，本研究所設計之介面在實務上的確能輔助專家學者進行文本相關研究，也能 讓對文本熟悉程度不一的使用者在利用此介面探索文本的過程中，更快速地掌握大規模文本的事件全貌。 / With the explosion of text information, there are more and more data being recorded and transmitted in the form of texts. However, as the amount of textual information becomes larger, how to effectively and efficiently realize the information also becomes more difficult. This study attempts to use the Topics Models, text-mining techniques to identify the important topics in the large textual information. In addition, this study also aims to use the techniques of data visualization to present the most informative and valuable details within the large texts. There are two parts in this work: the first part is the introduction of text mining algorithms and the second part is the design of the data visualization.Moreover, in the experiments, we also conduct several surveys to verify the proficiency and usefulness and the visualization design. The results of the experiments and surveys, supports that our design provides an effective and efficient interface for users to understand a large set of texts, even for the experts familiar with the corpus.

本論文研究目的，一為探討《蘋果日報》在台發行前，《中國時報》與《聯合報》在版面視覺元素的差異；二為《蘋果日報》在台發行後，三報在版面視覺元素方面的差異，以及在地兩報前後階段是否有差異。目的為了解在地兩大報是否因為《蘋果日報》加入競爭市場，而改變版面呈現。本研究採內容分析法，以次數分析、ANOVA檢定、雪菲檢定等為主。研究結果發現，《蘋果日報》的版面風格十分接近國外的小報風格，但《中國時報》與《聯合報》並未徹底顛覆過去報紙版面風格的定位。

在現今的社會中，我們可以隨處見到影像被用到各個地方，像是報章雜誌、網站或是兒童圖畫書中，影像可以加深讀者對文字的印象。對一般人而言，這些影像往往比周遭的文字來的更吸引人。尤其，童話故事的文本若以影像的視覺方式呈現，將更可吸引兒童的注意。 因此，本論文研究將文字形式的童話故事文本轉換為影像的視覺化技術。我們利用童話故事的敘事結構、角色等特性，將童話故事依故事劇情分段。從中找出代表每個段落主題與故事全文的關鍵字，並利用全球資訊網的影像搜尋引擎來找出初步的影像集合。最後再為每一段落找尋適合的影像，進而達到視覺化的效果。實驗結果顯示，本研究所提出的視覺化技術，在還原童話故事的敘事結構上，準確率約70%。 / Stories are often accompanied with images, in order to emphasize the effect of stories. In particular, most fairy tales written for children are decorated by images to attract children’s interest. This thesis focuses on story visualization technology which transforms text of a fairy tale into s series of visual images. The proposed visualization technology is developed based on the narrative structure of fairy tales. First, the input fairy tale is divided into segments in accordance with the plot of the story. Then, global keywords for the whole story and segment keywords for each segment are extracted. Moreover, expanded keywords which are important but infrequent in each segment are discovered. These three types of keywords are fed into Web Image Search Engine to find the initial image set. At last, the proposed system filters out the irrelevant images from the initial image set, and selects the representative image for each segment. Experiments show that the proposed method achieves 70% accuracy for the reconstruction of narrative structures of fairy tales.

如何透過視覺化探索勢力消長情形，是近年來頻繁被探討的問題，常見之做法會針對帶有時間屬性的時間關聯資料 (time-oriented data)來進行觀察，而以社群媒體為例，重大議題通常是透過意見領袖提出具有關鍵性之觀點，而得以分歧出新議題並吸引其他社群媒體上之閱聽人加入討論，上述之過程牽涉評論之階層資料其層次隨著時間變化分歧與合併，然而，能夠透過視覺化之方式同時觀察上述特性有其挑戰性。本篇論文將針對階層式資料提出一套整合方式，稱為TopicWave，特別是帶有時間變化屬性的資料，希望透過改良動態圖形視覺化工具，結合 Sunburst 與 ThemeRiver Graph，實作 Facebook 上公開文章之評論(comments)行為隨時間變化的趨勢，而透過直覺式互動功能之設計。透過案例分析和使用者測試，本論文提出的方法能清楚呈現評論關係隨時間之變化與階層式結構，達到組合式創新之效果。 / In recent research, it is a frequently asked question about how to explore the topic trend during a time interval. If we want to analysis and discuss this question, time-oriented data will be the most appropriate dataset. For example, on social media platform, major issues are commonly formed by opinion leaders, people will be attracted by opinion leaders and join in the commentary on a topic. The above-mentioned procedure will involve in commentary hierarchy level increasing or decreasing while time changes, however, it is challenging when we want to explore these properties using traditional visualization techniques. We propose TopicWave, a visualization design that combines ThemeRiver Graph (time-oriented visualization) and Sunburst (hierarchical data visualization). It can visualize the trend of a post’s comment on Facebook Page. TopicWave can clearly present hierarchy and time-varying trend of a Facebook post’s comment data at the same time through the intuitive design of interactive on visualization.

本篇論文針對社群使用者在社群網路文章上的互動行為進行研究，以視覺化工具Shareflow探索互動過程中造成資訊擴散的意見領袖以及傳播路徑。本研究主要分成兩個部分，第一部分為單一篇文章的分享路徑視覺化，基於階層化邊線綑綁(Hierarchical edge bundles)方法，根據控制點的引導將鄰近邊線進行綑綁，透過邊線捆綁舒緩資料量過大時造成之視覺混亂(visual clutter)問題。第二部分為粉絲專頁文章視覺化，分析多篇文章中具有多次分享、留言行為之使用者，呈現整體社群中積極活動的使用者以及其相關文章的視覺化。最後提供即時互動操作介面，以並列方式呈現出資料的廣度和深度，本研究的貢獻為提供一套視覺化工具，協助使用者探索臉書社群網路中的資訊散播過程以及發掘積極活動的臉書使用者。 / In this thesis, we propose a visualization tool, “Shareflow”, for observing the user activities in social media posts, and to explore opinion leaders and the propagation path caused by the information diffusion. Our approach contains two parts. The first part is a visualization of propagation path for a post in Facebook fanpage. Based on hierarchical edge bundles, we optimize the layout to reduce visual clutter caused by excessive information. The second part is visualization for a summary of posts in Facebook fanpage. It provides a tool for analysis the active users through their sharing and comments activities; In addition, we provide a real-time interactive interface, which demonstrates the breadth and depth of information concurrently.

打卡(Check-in)是Facebook平台上使用者經常使用的功能之一。過去關於打卡的研究大多採用質化的方法，而質性研究者在訪談使用者之前，為了解使用者的資料，往往需要手動查看使用者的Facebook塗鴉牆，收集、整理資料，費時費力。另外，使用者在Facebook上的打卡方式具多樣變化，許多的打卡不具有即時性與適地性。為了了解使用者的打卡動機，我們尚需取得使用者在手機上的操作與位置等資訊，方能還原打卡時的情境。為了了解打卡研究者的研究歷程與需求，我們讓質化研究先行，以進行需求分析，再依其開發一個協助研究者的分析工具，收集整理來自Facebook打卡資料與手機Log的資料，透過視覺化與列表的方式呈現，提供研究者能快速、深入分析、探索使用者在Facebook上打卡的行為與動機。本實驗邀請5位受試者扮演打卡研究分析者的角色，透過系統教學讓受試者學習使用系統，最後讓受試者自由探索使用者的資料集，並記錄下探索歷程與發現。實驗的評分結果使用5分量表，有用性向度的平均分數為4.6，受試者認為本系統能協助他們分析使用者的打卡行為與進行後續研究；易用性向度的平均分數為4.1，系統的操作方式對有的受試者需要時間來學習，但大部份受試者仍對本系統的易用性表示同意，證明本系統兼具有用性與易用性。另外，我們也發現受試者在探索過程與探索結果中展現了對此系統的創用性，是在設計者的預期之外，可見本系統的工具本質。 / Check-in is one of the functions that users often use on social media systems such as Facebook. In the past, most of the researchers about check-in use qualitative methods. Before interviewing users, qualitative researchers often need to manually check the user's timeline on Facebook to collect and sort out data, which is time consuming. In addition, the ways of a user's check-in’s on Facebook are diversified. Many check-in’s do not synchronize in time and place. To understand the motivation of a user in doing check-in’s, we need to collect more data such as the actions and location on user's mobile phone to restore the context. In order to assess the need of researchers, we use an interdisciplinary approach, in which qualitative research and tool development run in parallel. We develop a visual analysis tool aiming to help qualitative researchers to analysis data. Our tool collects check-in data from Facebook and synchronizes them with the user log on the user’s mobile phone. By visualizing the data in various forms such as bubble chart, timeline, map, and table, we hope to provide researchers with a quick overview of user behavior as well as in-depth studies of specific events. We use a 5-point scale to evaluate the system in our experiment. The average score of usefulness is 4.6, and the subjects think that our tool is very helpful for them to analyze check-in behaviors and the follow-up qualitative study. The average score of ease-of-use is 4.1. Although most of subjects agree that our system is easy to use, some think that practice is necessary to master the system. In addition, some subjects have found creative uses of our system that were not thought of by the designer, which reflects the nature of our tool in data exploration.

傳統的裝飾紋路，如獸面紋和雲紋，是經過工匠藝術家將其對於自然的變化與鳥獸的形態加以抽象化，藉由這些精妙的花紋傳遞豐富的訊息，形成具有象徵意義的紋路。本論文針對傳統中國雲紋進行分析，發現其保有高度的自我相似性，以此特性為基礎，我們發現到L-system也常被運用於此類特性紋路之創建，本論文以此語法為基礎撰寫生長規則，配合parametric L-system參數化的特性，不但可以讓雲紋達到很好的多樣性，同時配合非相片寫實電腦繪圖技術強化風格效果，讓我們的結果更忠於傳統雲紋，更進一步透過具時間序列的向量場資訊影響L-system之成長，產生具東方風格之動態雲紋。 / Traditional decorative patterns, such as animal mask and cloud patterns, are the art heritage from the abstraction of the changing nature and the shape of animals. These decorative patterns are not only exquisite art, but also meaningful symbolic patterns. In this thesis, we analysis the traditional Chinese cloud patterns, and find out the structure of self-similarity in cloud pattern. In this thesis, we design the growth rules base on the grammar of L-system and the parameterization of parametric L-system. We also combine the stylized line drawing to achieve a good variety of cloud patterns, and our results are faithful to the traditional art. Furthermore, we can generate the dynamic oriental cloud patterns with a time-varying vector field to affect the growth rule.

隨著社群媒體的普及，群眾產製的內容（User-generated content, UGC）時常成為新聞記者取材的對象，但現今隨著社群媒體爆發的資料量，記者不易從資料中看到事件的全貌，僅將社群媒體當作一種消息來源，因此報導的內容經常抄襲網友的意見或是落入片面討論的窠臼，無法駕馭社群媒體帶來的豐富資料。考慮改善這樣的現象，本研究透過將新聞取材的過程分為探索事件、收集素材以及回溯情境三個動作來協助記者探索新聞主題。以推特（Twitter）的資料為例，以網路為系統平台，開發一個輔助記者探索社群媒體上的事件、挖掘新聞主題的資訊系統，利用網絡分析以及自然語言處理的技術，結合視覺化的介面將事件資料集用故事元素的方式呈現，四種故事元素模型提供不同的觀察資料集的角度，並利用調整四種故事元素的權重，還原推文文本的語境，找出使用者想看的內容。我們設計了兩階段的任務式實驗以及評估問卷來證明系統的可用性，透過實驗結果驗證了本研究在以社群媒體輔助記者探索新聞主題的系統之價值，能讓對事件不同熟悉程度的傳播記者在此平台上探索新聞主題，並寫下深度報導的編採線索或是一篇新聞報導，透過本系統的輔助，讓使用者在探索及追蹤一起事件時，變得較為快速。 / With the popularity of social media, news reporters usually draw the news materials from mass user-generated content. However, with the outbreak of social media data, the reporter is not easy to see from the data in the whole picture of event. They only use the social media as a news source, so the reported content often copied the views of users, or fall into the stereotype of a one-sided discussion. The reporters can not control the wealth of information brought from social media. Consider improving this phenomenon, our study use Twitter data for example, develop an information system to assist reporters to explore the events on social media, and mine the news topics. We use network analysis and natural language processing as our technique, and show the story elements with the visualization interface. We apply four different story elements model, support the different way to explore data, and let user can adjust the weights from different model to retrospect to the context of tweets, help user find the news topics. We have designed a two-stage task experiment and assessment questionnaire to prove the availability of the system through experimental results. We can allow the reporters who are varying degrees of familiarity of the event to explore news topics from our system. We make the reporter to explore and track some events faster.

隨著社群媒體的普及，新聞媒體與意見領袖逐漸重視在社群媒體上以貼文方式發佈新聞資訊，社群媒體成為許多使用者會接收新聞與重大事件的主要管道且透過社群媒體的評論、分享與按讚等互動機制表達立場，這些即時互動行為是傳統媒體缺乏的機制，如何分析也是研究上的挑戰。本研究將針對Facebook上的貼文與互動行為進行分析，提供一款互動視覺化系統，找出貼文資料集中相似的貼文群集以及隨著時間推移下貼文屬性的變化，進一步瞭解Facebook上使用者、貼文與重大事件之間的相互影響。由於Facebook上的貼文與互動行為具多維度屬性，我們透過降維演算法將大量的貼文以二維散佈圖呈現，達到將相似貼文分群的效果。另外，我們設計了一種視覺化呈現方法，「Time Block」，突顯出時間的推移下貼文屬性的變化，藉此觀察出貼文資料集是否存在特定的模式。最後提供即時互動的操作介面，以及貼文屬性以及關鍵字兩者的統計，藉此連接到貼文集的屬性與時間的分佈關係，協助以視覺化方式進行探索與分析。最後，透過案例分析與使用者測試呈現此視覺化探索工具的優缺點。 / Social media becomes an essential medium for broadcast news. News media and option leader post information and people love to receive news and interactive using comment and likes to feedback. It is a research challenge to analysis this massive amount interactive behavior data in social media. In this paper, we propose an interactive visualization system to explore on the posts and interactions on Facebook. This system can help a user to find out the similar interactive behavior cluster and the trend of time-varying attributes to understand how the users, posts, and a big event to affect each other. Facebook Posts and interactive behavior contains multiple dimensional attributes; we adopt the dimensional reduce algorithm and 2D scatter plot to present the cluster in the spatial domain. Then, we design a time-varying visualization method, `Time Block,' can highlight the changing attributes and observe the unique pattern in the time domain. Also, we design a real-time interactive interface to connect the cluster and trend visualization with additional keyword distribution and attribute statistics. Finally, we use case study and user study to demonstrate the advantage of the proposed system.

每一項學術研究進行，其理論基礎都必需要建立於過去已完成的研究之上，因此文獻尋找與探討是進行研究過程非常重要的一個步驟。在數位時代與網際網路的加乘效益之下，改變了過去研究者必需為參考文獻東奔西跑的文獻資料尋找方式，但是卻會造成研究者被許多數位文獻淹沒。借用自網頁分析技術而設計的Google學術搜尋網路工具，能透過已經計算好的文獻權重PaperRank排序使用者所尋找的文獻集合，讓使用者能在數位文獻之中依單篇文獻被引用次數為原則而理出頭緒，但其順序式的排列仍然不能夠揭露出搜尋到的文獻集合裡彼此之間的關聯，其中包括了文獻所使用的關鍵字、作者與參考文獻。為了處理了解文獻中多維度的複雜資料關聯，最好的方式還是依賴人類的視覺化資訊處理能力，特別是當資料量大並且需要在短時間內決策時。 此外使用在文獻分析研究中，學者們使用共同引用(co-citation)、共同作者(co-work)、共同作者引用(co-author)等分析方式，配合延伸自社會網路分析理論中的社會密度(social distance)、關聯層級(social degree)、群(clique)等參數概念，試將複雜的文獻資料有脈絡地按排供參考。僅管此是工作難以機械化且消耗時間的(Börner, Chen , Boyack, 2003)，但是卻能將某一特定領域的發展直覺地呈現出來，如此若能將這些分析方式配合視覺化的呈現，則研究學者便能更進一步了進行大量文獻資料視覺化的分析、探索。 本研究試提出一個新的協助文獻探索平台系統架構，將傳統的文字搜尋轉變為視覺化的資料探索。使用者能透過三種不同的層級的資料：知識本體與關鍵字層、引文網路層及人員網路層，並與呈現的資料互動進一步了解資料間的關聯方式。最後實作視覺化雛型平台，並使用在國家圖書館所提供的博、碩士論文網所提供的論文資料，提供給研究人員探索特定知識領域中新研究方向的探索工具，並能協助研究者能在尚未完瞭解的專業領域之前，能快速地瞭解在該其領域重要文獻的導引平台。 / Paper survey is the most important task for building earnest theories, while researchers conducting academic researches. One must touches the fundamental detail of each theory and track down the develop-path of what achievement have been established by previous researches. Benefit from synergy of information age and document digitalized, it not only reduces the cost of finding reference documents, but also makes researchers suffer from information overwhelming after click single “search it” bottom. Stand in for traditional paper web search methods, new academic paper search technology borrowing from the idea of web search engine calculates the importance of each paper by cited number, and recommends users the most important papers by serial listing. However, serial listing does never spell the relationships of suggesting papers out, but only those results match some specific criteria. Those relationships of papers can be classified into 3 different types: the relations of keywords and references that author used and social relationship of authors like co-author and author co-citation which have been developed to explain the complex citation network structures. Those multi-dimensional relationships are extremely abundant and complex, so there is no better way to deal with but depending on visual data processing within human nature. In this paper, we try to propose a new platform to transform paper search in serial listing, into a visualized explore platform by demonstrating 3 different types of relationship: ontology-keywords, papers-references and personnel-references. End users can fallow the relationships between each difference nodes to explore considerable references, as well as change into different view and interact with existing information by using interactive mechanizes. In order to bring this idea to practical application usage, we build a proto-type platform to show our idea by using data from ETDS (electronic theses and dissertations system) of Ministry of education. We hope sincerely by using this proto-type platform, users can catch the major ideas of specific knowledge domain and researchers can explore acceptable references and even conduct new search topic.