基於智慧型手機應用程式及軌跡探勘個人活動區域

Abstract

探勘個人活動區域及其對使用者之意義是為個人化推薦及地理位置服務相當重要之關鍵，有眾多研究基於使用者軌跡探勘使用者活動區域，卻只考慮其區域於地理資訊上之意義，未考慮對於個別使用者有何個人意義。在同一區域內不同使用者會有不同的行為，而此區域對於使用者之意義需反映使用者於此區域之行為偏好。由於智慧型手機上行動應用程式（簡稱App）迅速發展以滿足使用者的於不同情況下需求，基於使用者使用App之行為模式，我們得以自動偵測使用者偏好。在本文中，基於使用者軌跡及行動應用程式之使用行為，我們將探勘個人活動區域並自動偵測此區域對使用者之意義。首先，由使用者軌跡於時間空間上密集程度，我們將找出使用者常停留之活動區域。而此區域對使用者之個人意義將可由地理資訊及使用者使用App之行為模式而得知。從使用者於區域內停留時所使用之App記錄可推論於區域內代表使用者行為偏好之App（representative App），而使用者使用App之行為模式便是由representative App概括總結而來。於此研究中，我們實際從使用者使用智慧型手機的紀錄進行綜合實驗。結果表明，此篇研究方法可以有效探勘個人活動區域及其對使用者之意義。

Discovering personal semantic regions that referring user preferences on locations is a crucial prerequisite for personal recommendation, location-based service, or data cleaning. A considerable amount of research effort has been elaborated on mining semantic meanings from user trajectories. However, previous studies discover semantics of regions only from geographic information provided. Given a region, since each user may have different kinds of activities within the region, the personal semantic of this region should reflect individual user’s intension. Due to rapidly increasing amount of mobile applications on smart phones, in which various applications could be executed assist user’s activities in different situations, making it possible to automatically identify individual user’s preferences from application usage patterns. Thus, in this paper, we propose a personal semantic region discovery framework based on users trajectories and application usage logs from smart phones. Explicitly, we explore density concepts in spatial and temporal domains to discover regions of interest from user trajectories. Then, the semantics of the discovered regions are identified by both geographic information and application usage patterns. The application usage patterns discovered from application usage logs summarize the representative applications, which reveal the user’s behavior preferences on the regions. In this study, we conduct a comprehensive experimental study on real datasets collected in smart phones and trajectory community. The results demonstrate the effectiveness of our approach to discover personal semantic regions.