地球信息科学学报 ›› 2020, Vol. 22 ›› Issue (4): 842-856.doi: 10.12082/dqxxkx.2020.190257

• • 上一篇    下一篇

语义轨迹建模与挖掘研究进展

赵斌1, 韩晶晶1, 史覃覃1, 吉根林1,*(), 刘信陶2, 俞肇元3   

  1. 1. 南京师范大学计算机科学与技术学院,南京 210023
    2. 香港理工大学土地测量及地理资讯学系,香港 999077
    3. 南京师范大学地理科学学院,南京 210023
  • 收稿日期:2019-05-27 修回日期:2019-10-31 出版日期:2020-04-25 发布日期:2020-06-10
  • 通讯作者: 吉根林 E-mail:glji@njnu.edu.cn
  • 作者简介:赵 斌(1978— ),男,江苏南京人,博士,副教授,主要从事时空数据挖掘研究。E-mail:zhaobin@njnu.edu.cn
  • 基金资助:
    国家重点研发计划项目(2017YFB0503500);国家自然科学基金项目(41971343);国家自然科学基金项目(41571379)

Advancements in Semantic Trajectory Modelling and Mining

ZHAO Bin1, HAN Jingjing1, SHI Tantan1, JI Genlin1,*(), LIU Xintao2, YU Zhaoyuan3   

  1. 1. Nanjing Normal University, School of Computer Science and Technology, Nanjing 210023, China
    2. The Hong Kong Polytechnic University, Department of Land Surveying and Geo-informatics, Hong Kong 999077, China
    3. Nanjing Normal University, School of Geography, Nanjing 210023, China
  • Received:2019-05-27 Revised:2019-10-31 Online:2020-04-25 Published:2020-06-10
  • Contact: JI Genlin E-mail:glji@njnu.edu.cn
  • Supported by:
    National Key Research and Development Program of China(2017YFB0503500);National Natural Science Foundation of China(41971343);National Natural Science Foundation of China(41571379)

摘要:

语义轨迹是时空轨迹和语义信息融合的产物。除了含有时空信息以外,语义轨迹包括移动对象自身的运动状态(如速度、方向)、环境(如气温、空间拓扑关系)和社交关系(如好友关系、社交活动)等多方面信息。挖掘语义轨迹可以深入地发现个体或群体移动行为的意图、习惯、情感等高阶语义内容,从而深层次发现个体或群体移动行为的模式、关系和规律等。因而,相较于时空轨迹,语义轨迹在语义性、解释性、可行性等方面更符合决策分析应用的实践需求,具有更重要的研究意义和应用价值。本文对语义轨迹挖掘的关键技术进行了综述。首先,介绍语义轨迹的基本概念,并且根据语义元素类型的不同总结了4种常见的定义形式。其次,归纳了语义轨迹建模的基本阶段,包括预处理、轨迹分段和语义富化。由于语义轨迹无法像时空轨迹那样从位置感知设备中采集获得,因此语义轨迹是通过建模技术得到的,主要通过将语义信息和时空轨迹相融合生成相应的语义轨迹。然后,介绍语义轨迹挖掘的主要任务,包括语义轨迹模式挖掘、语义轨迹聚类、语义轨迹分类、语义轨迹异常检测等。针对每一项挖掘任务,介绍了有关的基本原理和相关算法,总结了主要的关键技术和挑战。最后,探讨了语义轨迹挖掘现存的研究难点和未来研究方向。从模型定义、语义标注技术、多源数据建模等方面,讨论了语义轨迹建模的重要研究问题;从语义轨迹数据管理、分类和预测、流式数据挖掘、隐私保护、多粒度挖掘、评价方法等方面,探讨了语义轨迹挖掘的未来研究问题。

关键词: 语义轨迹, 语义轨迹建模, 语义轨迹挖掘, 语义富化, 本体知识库, 语义轨迹模式挖掘, 语义轨迹聚类, 语义轨迹分类, 语义轨迹异常检测

Abstract:

A semantic trajectory is a combination of a spatiotemporal trajectory and semantic information. Besides spatiotemporal information, a semantic trajectory comprises movement states (e.g. speed, direction), external contextual information (e.g. temperature, spatial topological relationships), and social relationships (e.g. friend relationships, social activities) of moving objects. We can derive from semantic trajectories intentions, habits, emotions, and other high order semantic information, thus further discover the patterns, relationships, and rules of individual or collective mobility behaviors. Therefore, compared with spatiotemporal trajectories, semantic trajectories are more in line with the practical requirements of decision-making applications in terms of semantics, interpretation, feasibility, and so on. This paper reviews the key technologies of semantic trajectory mining. First, we introduce the concept of semantic trajectories, and summarize four classic types of semantic trajectory definitions according to semantic elements. Then, we introduce the main phases of semantic trajectory modeling, including preprocessing, trajectory segmentation, and semantic enrichment. Since semantic trajectories cannot be acquired from location-acquisition devices as spatiotemporal trajectories, semantic trajectories need to be obtained through modeling techniques. Thus, the basic idea is to combine spatiotemporal trajectories with semantic information to generate corresponding semantic trajectories. Next, we introduce the main tasks of semantic trajectory mining, including semantic trajectory pattern mining, semantic trajectory clustering, semantic trajectory classification, anomaly detection of semantic trajectories, and so on. For each mining task, this paper introduces the basic principles and related algorithms, and summarizes the main key technologies and challenges. Finally, this paper concludes with the existing challenges and promising research directions of semantic trajectory mining. Specifically, this paper discusses the important research issues of semantic trajectory modeling in aspects including modeling definition, semantic annotation technologies, and multi-source data modeling. This paper also discusses the promising research issues of semantic trajectory mining such as semantic trajectory data management, classification and prediction, trajectory stream mining, privacy protection, multi-granularity mining, and evaluation methods.

Key words: semantic trajectory, semantic trajectory modeling, semantic trajectory mining, semantic enrichment, ontology knowledge base, semantic trajectory pattern mining, semantic trajectory clustering, semantic trajectory classification, semantic trajectory outlier detection