由於現今LBSNs (Location-Based Social Network) 的盛行，有越來越多POI(Point of Interest)相關服務來協助預測使用者可能有興趣的POI。在複雜的順序資料中尋找規律並不是件容易的任務，從先前的傳統的方法主要是利用POI之間的相關性來進行推薦。然而近年來也有許多研究利用深度學習的方法來做POI的預測，透過訓練模型來分析使用者的移動習性，在旅遊規劃時，地點的類別是一極大的影響因素，但較少研究著重於地點類別對POI預測的影響。本文提出了一新穎的POI推薦系統，以深度學習中的序列型模型(Sequence-to-Sequence)為基礎，進一步導入類型演變的觀念，分析了使用者目前的軌跡並預測一系列未來有興趣之地點。除此之外，本文亦提出C-S2S、DEC-S2S和IEC-S2S這三種新的學習模型利用地點的類別來提高預測的精準度。而實驗結果顯示，S2S確實能比傳統遞迴神經網路更能有效地利用序列間的關係做預測，而C-S2S、DEC-S2S和IEC-S2S也更提高了預測的精準度。;Owing to the great advances of mobility technique, more and more POI (point of interests)-related series have emerged, which could help user to navigate or predict the POI that may be interesting. Obviously, predicting POI is a challenging task; the complex sequential transition regularities, and the heterogeneity and sparsity of the collected trajectory data really hinder recommending the precise POIs. Prior studies of successive POI recommendation only focus on modeling the correlation among POIs based on users′ check-in data, while omitting the other feature of check-in data. Both the historical footprint of users’ check-in location and the type of location are important factors which influencing users’ decisions. We also take the category of location into consideration with different methods C-S2S, IEC-S2S and DEC-S2S to get more precise. The result also shows that S2S can capture the structure between sequence efficiently. The C-S2S model and IEC-S2S model also increasing the precision score.
