基于轨迹大数据的热点路径识别与查询处理关键技术研究

The arrival of big trajectory data is changing the management of spatio-temporal datasets in many ways. As a novel path-finding query, finding the most frequent path (MFP) plays a crucial role in many real-world applications like urban planning, spatio-temporal data mining, and various location-based services. However, finding MFP is challenging due to three reasons. First, it is nontrivial to give a satisfactory definition of MFP. Second, existing data access methods become suboptimal when querying trajectory data. Finally, the accuracy of the real-time approximate algorithms for MFP finding tends to be poor. To address these issues, we will study how to extract key properties of MFP and how to reasonably identify the MFPs. The goal is to try to reveal the common routing preferences of the past travelers. Based on this, we proceed to investigate the querying methods for different MFP queries. Specifically, we first study the computation functions for "path frequency" and its ranking systems. Then we study the queries of finding time period-based most frequent paths and constraint-aware most frequent paths in big trajectory data, respectively. Finally, we conduct extensive experiments using real datasets to evaluate the effectiveness and the efficiency of our proposed approaches.

轨迹大数据的到来对传统的时空数据查询与管理提出了新的机遇与挑战。作为一种新型的路径查找技术，热点路径的识别与查询已成为城市规划、时空数据挖掘，以及各种基于位置的服务等应用的关键核心技术。然而它也面临着路径的"热度"主观性强，形式化定义难度大；轨迹数据索引机制不完整，查询速度慢；以及查询复杂，实时算法准确差等诸多难题与挑战。 针对这些问题，本项目将以尽量准确的反映大众寻径的经验和智慧为目的，分析和提取热点路径的关键特征，设计合理的最优路径识别技术。并以此为基础，研究各类热点路径的查询与优化。我们首先研究路径的热度计算方法与排名机制，然后分别研究基于时段的和带条件约束的热度路径查询方法。最后我们将搭建系统和测试平台，利用真实的轨迹大数据进行实验，验证所提方法与技术的有效性和查询效率。

轨迹大数据和社交网络大数据的到来对传统的时空数据查询与管理提出了新的机遇与挑战。热点路径规划和社交网络中各类事务规划，已成为智慧城市以及各类时空数据挖掘等应用的关键核心技术。然而，它们也面临着难存储、难通信、难查询、难规划等诸多难题与挑战。. 针对这些问题，本课题在过去三年中分别对轨迹大数据和社交网络大数据进行了研究。轨迹大数据部分的研究内容包括：受道路网络约束的轨迹大数据压缩技术的综述、分类与比较；新型轨迹压缩技术的研究；受道路网络约束的轨迹大数据压缩体系框架的研究；基于轨迹大数据路由决策建模的驾驶经验学习；利用众包地图服务的道路类型预测；轨迹大数据异构数据副本存储结构设计与实现。社交网络大数据方面的研究内容包括：能效可知的社交网络事务及参与者规划；瓶颈可知的社交网络事务规划；基于分布式概率数据集的高频数据追踪；基于相互关联的不确定数据集的高频数据挖掘。到目前为止，课题已顺利完成研究任务和指标。发表学术论文9篇，其中CCF人机交互与普适计算A类国际会议Ubicomp 1篇，CCF数据库A类国际会议SIGMOD 1篇。课题组搭建了名为AVATAR的轨迹数据处理与查询平台，并在此基础上实现并测试了轨迹数据的压缩、存储、道路类型预测等技术。此外，课题组申请专利3项，其中1项已获批通过。参加国际会议并交流项目研究成果3次。