隨著近年來無線設備與技術上的進步,適地性服務(location-based services)中的地標興趣點(points of interest)推薦系統得到越來越多使用者的歡迎。這些系統通常採用協力式過濾(collaborative filtering)以達成個人化的推薦結果,並利用空間的資訊以改進推薦的品質。然而,除了使用者所在的地點外,時間的因素也深深地影響使用者的想法與喜好,因此,我們認為時間的資訊亦有助於改善推薦的品質。在這篇論文中,我們提出一個新的方式,將資料依時空的叢集進行分割,進而達成更正確且有效率的協力式過濾。在依據時空的資料分割下,使用者於不同時空環境中的想法與喜好可以被更有效地顯露與描繪出來。經過了許多實驗的驗證後,我們提出的PStarπ (Personalized Space-Time-Aware Recommender for Points of Interest)系統能夠提供更精確且有效率的推薦,並且更能滿足使用者的需求。
Recommendation systems for points of interest on location-based services have gotten widely popular as mobile devices and techniques progress in recent years. These systems usually use collaborative filtering approaches to achieve personalized recommendation and the spatial information is employed as an important factor inside to assist in improving recommendation quality. However, aside from users' locations, time also highly affects where users plan to go or what users prefer to see on location-based services, and the temporal information can benefit recommendation results very well. In this thesis, we propose a novel approach to partition data by spatiotemporal clustering and to utilize collaborative filtering with much better scalability and accuracy. The characteristics of users' preferences under different spatiotemporal situations will be revealed and preserved simultaneously under spatiotemporal data partitioning. Experiments show the recommendation quality provided by our PStarπ (Personalized Space-Time-Aware Recommender for Points of Interest) system is much more efficient, precise as well as satisfactory to users' needs.