請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/52383
標題: | 防災簡報資料之自動標籤 Automatic Tagging for Disaster-related Presentation Files |
作者: | Shiang-Wen Yang 楊翔文 |
指導教授: | 康仕仲(Shih-Chung Kang) |
關鍵字: | 災害管理,資訊傳遞,簡報資料,標籤,自動化, Disaster Management,Information Delivery,Presentation Tool,Tag,Automatic, |
出版年 : | 2015 |
學位: | 碩士 |
摘要: | 防災管理需要資訊,近幾年防災簡報資料是由防災領域的專家製作並用於傳遞關鍵的資訊,而充足且快速的資訊傳遞是政府部門決策的關鍵,換言之,於災害管理(Disaster Management)中一個健全的資訊傳遞系統可以讓使用者有效率得獲得適合的資訊。在眾多災害資料類別中,多媒體資料(如簡報資料)鮮少被討論和利用,如沒有適當的資訊萃取(Information Extraction)及資訊取得(Information Retrieval)程序,該資訊將無法加以活化,進而轉化為知識以供傳遞。
本研究開發自動標籤法,其中包含四階段程序,文字萃取程序(Text Extraction Process)、詞彙擷取程序(Word Segment Process)、原生標籤程序(Feature Tagging Process)及衍生標籤程序(Reasoning Tagging Process),本研究亦提出一邏輯運算,透過此邏輯運算可找出現有資料庫與不同標籤間之連結,並以Rule-based和Document-centered的概念發展防災簡報資料的自動化標籤方法,將文字萃取程序及詞彙擷取程序的結果放入各種標籤類型的規則(Rule)中以產生標籤,其標籤可以分為三類:原生標籤(Feature Tag)、簡易衍生標籤(Direct Reasoning Tag)及複雜衍生標籤(Complex Reasoning Tag),簡易衍生標籤及複雜衍生標籤分別是利用一個或兩個現有標籤連結資料庫而得,利用標籤之間的資訊交疊,使防災簡報資料能夠快速得被分類、搜尋、運用,並可協助決策者進行判斷。 本研究進行效率(Efficiency)及效能(Performance)分析,效率分析挑選15份簡報,共253頁之內容計算各個程序所花費之計算時間,平均單一標籤產生時間為1.0秒;效能分析之測試資料是從前一份測試資料利用系統抽樣選出10個頁面,我們邀請兩位防災領域的專家,請他們以人工的方式選出標籤,再與本研究之結果比較,專家A選出了53個標籤,專家B選出150個標籤,兩位皆耗時約10分鐘,本研究產生29個標籤,耗時約1分鐘,我們分析系統與專家結果的差異,有30個專家A挑選的標籤及122個專家B挑選的標籤是在本研究之限制內,移除因限制未產出的標籤,專家A挑選出23個標籤,專家B挑選出28個標籤,本研究於改善限制後可以產生接近擁有專業知識的專家所提供的結果,且更快速。 本研究發展防災簡報資料的資訊萃取及資訊取得程序,此方法可以讓使用者在災害應變時快速取得適合的資訊,使防災簡報資料在災害管理中可以產生更大的應用價值。 Disaster management requires information. Presentation files were created from experts who specialize in disaster response and used to deliver critical information during disaster period in recent years. Sufficient and rapid information delivery is the key for government to make decision. That is, a sound information delivery system in disaster management let user acquire the suitable information efficiently. However, with several types of disaster data, multimedia data, such as presentation files, has few discussion and utility. In case of lack of proper information extraction (IE) and information retrieval (IR) process, the information can be applied on further application and convert into knowledge. In this research, an automatic tagging for disaster-related presentation files is developed, which includes four process, text extraction process, word segment process, feature tagging process and reasoning tagging process. We also raise a logic manipulation to find the connection with tags and database and implement this method by the concept of rule-based and document-centered. While the data completed text extraction process and word segment process, it would run though creation rules of each kinds of tag and three types of tags: feature tag, direct reasoning tag and complex reasoning tag are created. Direct reasoning tag and complex reasoning tag are produced by connecting one or two existing tag with database. With the interaction in tag and database, the disaster-related presentation files can be categorized, searched and utilized. This research conducted efficiency and performance test. We selected 15 pieces of presentation files (totally 253 pages) and computed the computing time in each process. The average time to produce one tag is one second on average. We also selected 10 pages of slides form previous data set by systematic sampling for performance test. We invited two experts in disaster prevention domain to tag those 10 pages and compared the result with tags created by our process. Expert A collected 53 tags in about 10 minutes while expert B found 150 tags in about 10 minutes. The new tagging technique tagged 29 tags within 1 minute. There are 30 tags of expert A result and 122 tags of expert B result under limitations our process can deal with. While ignoring the limitations, expert A finds 23 tags and expert B finds 28 tags. In short, our process can get result close to human with professional knowledge and much faster. This research develops information extraction and information retrieval process of presentation files. This technology can help user acquire the suitable information during the disaster period and make presentation files create much more value in disaster management. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/52383 |
全文授權: | 有償授權 |
顯示於系所單位: | 土木工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-104-1.pdf 目前未授權公開取用 | 5.62 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。