請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/77770
標題: | 災害應變問答對話系統 Question-Answering Dialogue System for Emergency Operations |
作者: | Hao-Yung Chan 詹皓詠 |
指導教授: | 康仕仲(Shih-Chung Kang) |
關鍵字: | 災難管理,災害應變,問答系統,對話系統,決策支援, Emergency Operations,Disaster Management,Dialogue System,Question-Answering System,Decision Support, |
出版年 : | 2018 |
學位: | 碩士 |
摘要: | 資訊量的快速增長,導致了決策的複雜度,並提升了決策者準確而有效地完成任務的難度,因此本研究試圖協助災害應變中心決策者,提高其利用資訊的有效性與準確性。在前期研究中,我們為決策者開發一個聊天機器人,以協助決策者更有效、準確地利用、選擇和處理資訊。經過使用者測試後,我們發現對話系統必須更加強在貼合災難管理和災害應變的特點和需求。因此,本研究開發了結合問答功能的對話系統,引入知識庫來提供資料與資訊,並納入搜索模組來處理相對困難的查詢任務,此外,為了提升災害管理對話系統的性能,本研究專注於發展語意理解技術,以提高產品分析和理解使用者需求的能力。在本研究中,問題的分析方法和知識庫都經過特殊設計,以符合災難管理的特性。從系統性能的驗證結果可以知道,本研究所開發的系統,在分析使用者問題、處理成機器可讀格式的表現,有百分之七十的成功率。 This research aims to eliminate the difficulties in efficiently and accurately utilizing information for the decision makers in the emergency operation center (EOC). The rapid growth of the amount of data causes the complication of the tasks of decision making in disaster management, and has increased the difficulties of the decision makers to accomplish their mission accurately and efficiently. In our preliminary work, we attempted developing a chatbot for decision makers and staffs of the EOC to assist them in utilizing, selecting, and processing information efficiently and accurately. The user experience test revealed that a better adaption of the dialogue system to the characteristics of disaster management and the EOC is required. Thus, we specifically focus on the development of the technique of language understanding for analyzing the user's demand. As a solution, we developed a dialogue system combining the feature of question answering, a knowledge base as the knowledge provider, and a module of searching to handle relatively difficult tasks of queries. To improve the performance of the system, we focus on enhancing the capability of understanding users' questions. The construction of question analysis and the knowledge base are both adapted to the characteristics of disaster management. The validation shows that our method can analyze and process the user's question into a machine-acceptable form with the success rate of about 70%. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/77770 |
DOI: | 10.6342/NTU201800945 |
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-107-R04521618-1.pdf 目前未授權公開取用 | 2.28 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。