III: Small: Semantic Trajectory Mining with Contexts

传感和定位技术的快速发展为我们提供了越来越多的从人类运动，动物痕迹和交通中收集的轨迹数据。理解这样的大规模轨迹数据连同它们的周围背景（例如，位置信息、当地事件、天气和环境）可以使许多重要的应用受益。例如，对人类轨迹的语义理解可以帮助分析一个人的兴趣，社会经济地位和健康状况;挖掘交通模式w.r.t.当地的事件和天气条件可以导致一个更有弹性的交通系统;研究动物运动如何对环境变化作出反应可以促进我们对生态系统的理解。本计画研究资料探勘演算法，并提供解决方案，以进行具有丰富时空背景之语意轨迹探勘。通过PI的跨学科合作，其成果将在社会科学、健康、交通和生态学等其他学科产生更广泛的影响。本项目旨在寻求轨迹数据挖掘的创新解决方案。现有的轨迹数据挖掘方法大都局限于只研究轨迹数据，而没有考虑轨迹数据丰富的时空背景信息。这样的限制常常导致检测到琐碎的甚至错误的模式。例如，异常（例如，如果考虑诸如大型足球比赛或极端天气之类的背景，则实际上可以预期交通拥挤或绕行轨迹）。要将上下文纳入轨迹挖掘，关键的挑战是在给定轨迹和上下文的稀疏和噪声观测的情况下，对上下文和轨迹之间的隐式和复杂的相关性进行建模。为了在分析轨迹数据时正确、充分地利用丰富的上下文，本项目有两个研究目标：（1）基于上下文的轨迹标注和（2）基于上下文的轨迹挖掘。在第一个目标中，目标是将轨迹与相关的静态或动态上下文相关联。在第二个目标中，研究人员将重新审视轨迹挖掘算法，并创新三种重要的挖掘技术：递归模式挖掘，异常检测和轨迹聚类。将开发新的数据挖掘技术，以便在数据挖掘过程中纳入带注释的上下文。该项目将推进轨迹数据挖掘的最新技术水平，丰富一般数据挖掘原则。

Rapid advances of sensing and positioning technologies have provided us with an increasing amount of trajectory data collected from human movements, animal traces, and traffic. Understanding such large-scale trajectory data together with their surrounding contexts (e.g., location information, local events, weather and environment) could benefit a number of important applications. For example, a semantic understanding of human trajectories can help profiling a person's interest, socioeconomic status and health conditions; mining traffic patterns w.r.t. local events and weather conditions can lead to a more resilient transportation system; and studying how animal movements respond to environmental changes can advance our understanding of the ecological system. This project investigates data mining algorithms and provides solutions toward semantic trajectory mining with rich spatial-temporal contexts. The results will have broader impacts in other disciplines such as social science, health, transportation, and ecology through PI's interdisciplinary collaborations.This project seeks innovative solutions for trajectory data mining. State-of-the-art trajectory data mining methods have been largely limited to studying only trajectory data without considering the rich spatial-temporal contextual information. Such a limitation often results in detecting trivial or even erroneous patterns. For example, anomalies (e.g., heavy traffic or detouring trajectories) may actually be expected if considering the contexts such as big football games or extreme weather. To incorporate contexts in trajectory mining, the key challenge is to model the implicit and complications correlations between contexts and trajectories given the sparse and noisy observations on both trajectories and contexts. To correctly and fully utilize the rich contexts in analyzing trajectory data, this project has two research aims: (1) trajectory annotation using contexts and (2) trajectory mining with contexts. In the first aim, the objective is to associate trajectories with relevant static or dynamic contexts. In the second aim, the researchers will re-visit trajectory mining algorithms and innovate three important mining techniques: recurrence pattern mining, anomaly detection, and trajectory clustering. New data mining techniques will be developed to incorporate the annotated contexts in the data mining process. The project will advance the state of the art in trajectory data mining and enrich general data mining principles.