以深度學習與週期性樣式探勘為基礎之使用者地點長期預測

Abstract

近年來，隨著行動通訊技術的進步和第四代行動網絡的發展和日益普及與位置定位技術的發展，行動通設備已經產生了有關人類、車輛、動物等大量的移動軌跡數據，反映出相關物體的移動性，許多新創團隊亦透過預測用戶的下一個位置提供了新穎的服務。目前現有的研究只能預測用戶的下一個位置，也就是是短期位置預測，但是卻無法運用於長期位置預測之情境。因此，本論文主旨為發展出一套長期位置預測的架構與方法。我們認為，如果我們能夠提高長期位置預測的可靠性，目前依賴短期位置預測的服務可以受益，甚至可誕生出更多創新且獨特的服務。在本文中，我們提出了一個基於深度學習和週期性樣式探勘的架構與方法進行長期位置預測。我們的預測架構與方法套用了自然語言模型的想法，並採用多步遞歸策略以進行長期預測。為了減少多步遞歸策略所的累積的誤差損失，我們利用週期性樣式探勘技術，減少所需的遞歸次數，進而減少損失，提高預測架構與方法的可靠性。基於真實世界的移動軌跡數據，我們進行了一系列的實驗。實驗結果顯示，本研究所提出的預測架構與方法可以做出有效的長期位置預測。

In recent years, with the advances in mobile communication and growing popularity of the fourth-generation mobile network along with the enhancement in location positioning techniques, mobile devices have generated extensive spatial trajectory data, which represent the mobility of moving objects. New services are emerged to serve mobile users based on their predicted locations. This thesis is concerned with the long-term location prediction of a user. Most of the existing studies on location prediction can only predict one next location of a user, which is regarded as short-term next location prediction, and they are not applicable for long-term location predictions. We believe that if we can improve the accuracy of long-term next location prediction predict, every current service that takes benefits of predictability on next location can be further extended. In this thesis, we propose a prediction framework named LSTM-PPM that utilises deep learning and periodic pattern mining for effective long-term location prediction. Our framework devises the ideology from natural language model and uses multi-step recursive strategy to perform long-term prediction. To reduce the accumulated loss in multi-step strategy, we utilise further the periodic pattern mining technique. Through empirical evaluation on a real-life trajectory data, our framework is shown to provide effective performance in long-term location prediction.