| 摘要: | 隨著無線通訊技術與普及運算環境發展，行動服務已成為行動商務中重要的應用領域。行動服務的特色是能夠依據使用者所在的位置，將適當的服務即時地提供給使用者，以達成日常生活便利化及工作流程加速化等不同的應用目的。然而，目前的行動服務技術是以使用者「現在」的位置作為服務驅動點，由於系統對該使用者過去的行動習慣以及未來可能的行動一無所知，使得系統無法針對個別的使用者，進行服務內容的個人化調整，造成行動服務的發展受到限制。因此，若系統能預測使用者未來的位置，並據以提供個人化的智慧型服務，將可有效拓展行動服務類型與價值。為突破目前行動服務發展的限制，行動服務系統必須具備學習使用者的行動習慣，並預測使用者未來行動位置的能力。由於行動服務應用的範疇包含室內與室外，目前室內/外定位技術皆已有相當程度的發展，如何設計高準確度的位置預測機制，並搭配適用的定位技術，分別設計室內及室外的使用者位置預測系統，以支援智慧型的服務應用，已成為重要的研究議題，因而引發本計畫的研究動機。本計畫提出具學習能力的行動使用者位置預測機制，此機制藉由分析使用者的移動歷史紀錄與最近的移動紀錄，從而學習使用者的移動樣式，據以預測使用者未來的位置。此外，研究將藉由設計動態推論資料庫，以期使預測機制在取得最新的移動紀錄時，可即時地對既有的移動樣式進行修正；換言之，當使用者的移動習慣改變時，系統可以在不須停機更新的狀況下，即時學習使用者的新習慣，並讓所支援的行動服務維持在線上運作的狀態，提供不間斷的智慧型服務。再者，根據文獻探討的結果，針對不同人的行為模式以及不同季的移動資料進行研究，五種傳統預測技術的平均準確度如後：Markov(81%)、State predictor(80%)、Bayesian network(83%)、MLP(77%)、以及Elman net(80%)，多只能達到八成左右的準確率。由於本定位機制能夠及時掌握使用者的移動習慣改變，故將能提供更為準確的預測結果，初步評估能將平均的準確度提高到九成以上。本研究提出具學習能力之使用者位置預測機制，未來完成後可以此機制為基礎，可建構室內或室外的位置預測系統，透過動態學習的特性，提供行動服務提供者低成本、易維護、且可針對個別使用者進行準確位置預測的平台，預期將能協助服務提供者開發更符合使用者需求的智慧型服務，藉以拓展行動服務的應用範疇與價值。 ; With the development of wireless communications and pervasive computing technologies, mobile services have become an important issue in mobile commerce area. The feature of mobile services is to provide appropriate services instantly for users according to their locations. However, the existing schemes of mobile services are designed based on current locations of users without knowing their moving habits in the past and possible movement in the future, which greatly limit the applications of mobile services. If the personalized mobile services can be given by predicting a user’s future location, the types and values of mobile services can be further advanced. To break the above limitations, mobile service systems must have abilities of both learning the moving habit and predicting the future location of each user. The localization technologies for mobile services are mature since nowadays services can be delivered indoor or outdoor. The motive of our research is to construct a location prediction mechanism which is associated with appropriate localization technologies to support smart mobile services. A location prediction mechanism with learning abilities for supporting mobile services will be proposed in this project. By analyzing historic and recent moving records of users, the proposed mechanism can learn their moving patterns so as to predict their future locations. Moreover, a dynamic inference database will be employed to update the moving pattern simultaneously according to the recent moving records of users. That is, while the moving habits are changed, the new habits can be obtained without shutting down the system. Hence, the feature will also keep the supported smart service system online to provide services continuously. Furthermore, five traditional prediction technologies were reported in surveyed research literature, which were applied to different users’ moving habits and various moving data of different seasons. The average accuracies of these prediction technologies are 81%(Markov), 80%(State predictor), 83%(Bayesian network), 77%(MLP), and 80%(Elman net). For the proposed scheme in the project can learn the changing of moving habits of users immediately, more accurate location prediction can be achieved. The average accuracy of the proposed mechanism is supposed to be higher than 90%. This project will propose a location prediction mechanism with learning abilities. Based on the proposed mechanism, the indoor and outdoor location prediction systems can be constructed. Through features of dynamic learning, the constructed systems can be obtained for the mobile service providers with low cost, easy maintenance, and accurate location prediction ability. The novel system can support the service providers not only to develop more services which can match users’ need, but also to expand the applications and values of mobile services. ; 研究期間 9808 ~ 9907 |
