時間擴展技術應用於動態遊程規劃：異常氣象資訊的規劃決策

Abstract

旅遊的本質即是體驗並旅行至人們日常生活空間以外的環境，而為了降低旅行者做出必要旅遊行程抉擇之資訊負擔，許多輔助遊客決策之旅遊推薦系統亦加入情境資訊以增進推薦之準確度。然而，由於遊客在多種異常天氣情境下之決策過程仍缺少實證研究支持，多數推薦系統之研究仍無法有效的將這些動態資訊結合於其系統內，僅顯示在介面上或是提供該資訊的連結而已。本研究以情境氣象資訊作為可能導致遊客改變或延遲旅遊行程的決定因子之代表，並對於遊客在特定情境下的決策之整體理解有所貢獻。最終透過旅程規劃之專家訪談與氣象指標之研究回顧提出相應之遊客決策概念模型。 利用交通部觀光局所提供之觀光資訊景點資料庫(共3,688筆)以及中央氣象局所提中之各景點氣象預報資料。本研究所建立之模型將應用於一行動智慧行程表系統，可透過即時資訊預警使用者可能影響行程之事件，並提供給使用者客製化的旅遊行程。同時為了使得該系統能夠介接隨時間變動之氣象及交通資訊，其行程推薦之演算法則是應用時間擴展網絡技術(Time-Expanded Network, TEN)以解決動態遊程規劃問題。研究結果證實TEN除了能夠在現實情境中找出最適旅遊行程外，亦能對於遊程中之不確定性突發事件做出調整，且亦可作為未來其他情境感知推薦系統之數模型基礎。

As the nature of tourism involves people traveling to places outside their usual environment, multiple travel recommender systems utilizing contextual information have been designed to reduce the information burden of a traveling individual making necessary decisions. However, due to the lack of empirical evidence on how tourists respond to various weather conditions they encounter, many existing applications have only partially implemented these information, merely providing direct access for users while not used as contextual data in their intelligent tourism systems. This current research contributes to the understanding of tourists’ decision making under certain travel contextual situations, as weather context is taken as a representative of determinants causing changes and delays in the individual’s trip. Further proposing a conceptual model through interviews with experts in tour-planning, paired with literature review of finding physical thresholds under varying weather information. Using the attractions data set of 3,688 POIs provided by the Taiwan Tourism Bureau and their corresponding weather forecasts retrieved from the Central Weather Bureau, Taiwan. Model valuation was done by designing a personal adaptive itinerary mobile application, taking weather forecast data into account and providing personalized travel itineraries accordingly. In order for the proposed decision support system to integrate time-dependent weather data and transit times, the implemented algorithm was built to be solved as a tour planning problem under its Time-Expanded Network (TEN). The results show that the TEN is highly capable of finding the most optimal travel itinerary and presenting unexpected event changes even in real-world scenarios, and may be further used as a mathematical model basis for other context-aware recommender systems.