中文關鍵字: 聊天機器人、決策系統、語意分析、擬人化腳本
英文關鍵字: chatting robot(chatbot), decision supporting system, semantics analysis, anthropomorphic script
中文摘要: 本計畫開發一對話式防災決策輔助系統(C-DSS),透過具人工智慧的防災對話機器人,於防災應變期間處理具有即時性、多樣性、不確定性之防災資訊。資訊量快速增長導致決策的複雜度,為因應大量且複雜之複合性災害,許多單位相繼開發整合各式防災資訊的決策支援平台。然而,現今平台多以資料整合為主,缺乏對話式的防災資訊提取及溝通機制,導致決策者須仰賴幕僚團隊大量人力,從既有平台蒐集資料並分析,方能做出對應決策。為解決此問題,本計畫開發對話式防災決策輔助系統,包含:(1)水利防災決策樹模組:根據防災資料需求,建置出一防災決策樹;本計畫經資料盤點後,建置包含169個節點的決策樹。(2)水利防災資料庫模組:整合各式政府開放資料以建置完整之防災資料庫;本計畫使用知識庫的設計方法,並開發防汛應變關鍵字資料庫。(3)防災語意分析模組:發展防災相關的語意分析機制,讓使用者能以自然語言索取所需資訊;本計畫基於問答系統原理研發本模組,針對防災資料時間特性,改良既有語意分析技術。(4)對話式應答模組:透過擬人化文字應答腳本與視覺畫圖像應答的設計,將防災資訊以更有效且直覺之方式呈現給決策者。各模組皆經特殊設計,以符合災難管理的特性。本計畫訪談了水利署防災中心人員與高階決策者,以了解使用者需求。經系統性能驗證,本計畫成果在分析使用者問題的表現,有百分之七十的成功率。本計畫與LINE及酷必資訊公司合作,共同進行整體系統開發及優化,亦在水利署災害緊急應變小組進行實測,以水利署107年水利災害應變服務計畫為實證場域,透過於汛期及颱風豪雨應變期間之實際操作,驗證系統有效性,並將計畫成果推廣至實際運作機制中。
英文摘要: This project aims to develop a Conversation-based Decision Support System (C-DSS). The rapid growth in the amount of data has caused complications in decision making tasks in disaster management. In countering massive and complex compounded disasters, many units develop all types of decision supporting platform for disaster data. However, the major functions of the current decision supporting platforms in the market are mostly for data integration while lacking intuitive disaster data acquiring and communication mechanisms; the decision makers still rely on teams of staff to gather data from the existing platforms and analyze before making countering decision. To solve this issue, this project developed a conversation-based system for decision supporting. The system includes, (1) water disaster decision tree module: a disaster decision tree including 169 nodes was established based on the needs of disaster prevention data. (2) Water disaster database module: a complete disaster database and a water-disaster-related key term database were established by integrating all types of government open data. (3) Disaster semantics analysis module: a semantics analysis mechanism based on the concept of question-answering systems concerning disaster prevention and temporal characteristic of disaster information was developed for the users to request for desired data with natural language. (4) Intuitive responding module: design with anthropomorphic text responding scripts and visualization imaging responding to present upper decision makers with disaster data in a more effective and intuitive manner. The staff and decision makers of the Water Resource Agency were interviewed to understand the user’s needs. The validation shows that the method can analyze and process the user's questions with a success rate of about 70%. This project was worked with LINE and CoolBe Co., Ltd. for platform development and optimization. For the project result testing, this project worked with Center for Weather and Climate Disaster Research at National Taiwan University and the Water Resource Agency Disaster Emergency Response Service of Ministry of Economic Affairs to conduct the field testing during emergency responses in flood season, typhoon and heavy rain. This perfected the C-DSS and promote the implementation effectively to the actual operation mechanism, allowing the upper-level decision making officers at the Water Resource Agency to use the system in supporting their disaster prevention decision making.