Elsevier

Information Sciences

Volume 502, October 2019, Pages 164-189
Information Sciences

Semantic periodic pattern mining from spatio-temporal trajectories

https://doi.org/10.1016/j.ins.2019.06.035Get rights and content

Abstract

Rapid development in GPS tracking techniques produces a large number of spatio-temporal trajectory data. The analysis of these data provides us with a new opportunity to discover behavioural patterns. Spatio-temporal periodic pattern mining is finding temporal regularities for interesting places. Mining periodic patterns from spatio-temporal trajectories reveals useful and important information about people’s regular and recurrent movements and behaviours. Existing periodic pattern mining algorithms suffer from two main drawbacks. They assume regularly sampled and evenly spaced trajectory data as input which is unlike real world data, traditional methods also fail to consider background aspatial information despite many applications requiring a semantic interpretation of movement behaviours. In this paper, we propose a new semantic periodic pattern mining algorithm from spatio-temporal trajectories that overcomes these two drawbacks from past studies. Experimental results with real world datasets demonstrate the efficiency and effectiveness of our proposed method.

Introduction

Due to the rapid development of positioning technologies such as GPS and GSM networks, spatio-temporal trajectories of mobile objects has become increasingly available. Such a large number of spatio-temporal trajectories provides us with new opportunities to discover behavioural patterns and periodic movements from the spatio-temporal context. Mining these spatio-temporal trajectories is one of the most popular data mining topics nowadays [12]. Periodic patterns reveal repeated behaviours that occur at regular time intervals in specific places. Mining spatio-temporal periodic patterns has attracted attention recently [20], [39], and there is a growing demand to analyse huge spatio-temporal trajectory data to find hidden and valuable periodic patterns to understand the repeating and regular behaviours of moving objects.

Recently, there has been some work in general periodic pattern mining from social networks [10] and from sequential data [11], [37]. In the spatio-temporal context, several attempts have been made to mine periodic patterns from spatio-temporal trajectories [7], [12], [18], [19], [20], [39]. These approaches try to take some of the special characteristics of spatio-temporal trajectories [38] into account when mining periodic patterns in order to find spatially and temporally meaningful periodic patterns. The spatio-temporal aspect in these approaches is either loosely-coupled [7], [12], [18], [19], [20] or tightly-coupled [39]. The former does not consider both spatial and temporal aspects equally and simultaneously, but follows a spatial-dominant approach where spatial aggregations are computed first while ignoring the temporal dimension, and then finds temporal regularities for these spatial concentrations. The latter considers both aspects at the same time in order to find spatio-temporal concentrations.

However, these past studies share two common drawbacks. First, they assume that GPS trajectories are regular trajectories, characterised by a constant time period between two successive recordings. That is, they assume that spatio-temporal trajectories are regularly sampled. In the real world, this is not always possible due to weather conditions, device malfunctions, battery issues, bandwidth limitations and power issues [16], [39]. Many spatio-temporal trajectories are irregular trajectories where variable time periods occur between two successive recordings. Therefore, these traditional approaches cannot be directly applied to spatio-temporal trajectories, but instead require a computationally expensive trajectory interpolation [3] to make the irregular spatio-temporal trajectories regular. Second, these approaches consider two important dimensions (spatial and temporal) for mining periodic patterns, but disregard another important aspatial semantic dimension (descriptive geographical feature information). Spatio-temporal trajectories are geographical phenomena occurring in the geographical space. Geographical phenomena have three dimensions: spatial, temporal and aspatial semantic [25], [36], thus considering the aspatial semantic dimension in spatio-temporal trajectory mining is of importance in order to not miss any spatially, temporally and semantically meaningful periodic patterns.

In this paper, we propose a semantic periodic pattern mining from spatio-temporal trajectories. We first detect spatially, temporally and semantically aggregated concentrations from irregular trajectories using aspatial semantic information from OpenStreetMap,1 and compute regular periodic time intervals for these spatio-temporal concentrations from irregular trajectories.

Main contributions of this paper are:

  • to propose a novel trajectory representation method describing a spatio-temporal trajectory as a sequence of semantic episodes that match background aspatial semantic information;

  • to discover spatially, temporally and semantically aggregated concentrations as reference spots for periodic pattern mining;

  • to detect regular time periods for each spatio-temporal concentration from irregular trajectories;

  • to provide experimental results to demonstrate the efficiency, effectiveness and applicability of the proposed method based on two real datasets.

The rest of paper is organised as follows. Section 2 briefly reviews previous spatio-temporal periodic pattern mining. Section 3 introduces our proposed semantic periodic pattern mining from irregular spatio-temporal trajectories. Section 4 describes two datasets used in this study, and presents a performance evaluation of trajectory interpolation that is required for existing approaches in periodic pattern mining. Section 5 reports comparative experimental results with existing approaches, and demonstrates the effectiveness and efficiency of our approach. Section 6 concludes and lists potential future directions.

Section snippets

Literature review

First, we formally define a spatio-temporal trajectory before we move to a literature review. Note that, the main focus of this paper is on periodic pattern mining from irregularly sampled spatio-temporal trajectories. Therefore, this section will mainly review existing periodic pattern mining approaches for spatio-temporal trajectories, and discuss their strengths and weaknesses with respect to the special characteristics of spatio-temporal trajectories to highlight the drawbacks of

Semantic spatio-temporal trajectory

A semantic spatio-temporal trajectory, Tsem, is a list of spatio-temporal, semantically annotated entries, (⟨x1, y1, t1, a1⟩, x2,y2,t2,a2,,xn,yn,tn,an), where xi, yi ∈ R2, tiR+, and aiA, for 1 ≤ i ≤ n and t1<t2<<tn. A is a finite set of semantic labels.

A semantic episode, Ej,k, is a single vector that represents a portion of movement from a semantic trajectory, Tsem. The portion of movement starts from the trajectory’s jth index and ends at its kth index (inclusive), where j ≤ k. Note

Datasets

Two real datasets are used for experimental studies in this paper. Due to the nonexistence of ground-truth dataset for periodic pattern mining evaluation, the first GPS dataset was intentionally collected by authors from 20/9/2017 to 20/10/2017 in Cairns, Australia. This generated a ground-truth benchmark dataset in order to validate and evaluate our proposed approach against past studies. There are two things to consider in this dataset. First, to measure the effectiveness of proposed

Efficiency

Fig. 12 displays the efficiency analysis of three approaches for Dataset 1 and Dataset 2. Although Periodica exhibits better efficiency than Traclus (ST), Periodica and Traclus (ST) are both inefficient, especially when 10 s is used as the time interval. It becomes computationally inefficient. Even when we use 120 s as the time interval, Periodica still spends 114.624196 s and 68.748305 s for Dataset 1 and Dataset 2, respectively. Note that, our method does not need an interpolation step, and

Conclusion

Mining periodic patterns from spatio-temporal trajectories is of great importance as it reveals interesting and regular periodic behaviours. There is growing interest in efficient and effective periodic pattern mining from spatio-temporal trajectories due to the wide availability of automatic location collecting devices. In this paper, we identify two crucial drawbacks of existing spatio-temporal periodic pattern mining approaches, Periodica and Traclus (ST): (1) inefficiency due to inevitable

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References (39)

  • A. Bohn et al.

    Identification of rhythmic subsystems in the circadian cycle of Crassulacean acid metabolism under thermoperiodic perturbations

    Biol. Chem.

    (2003)
  • H. Cao et al.

    Discovery of periodic patterns in spatiotemporal sequences

    IEEE Trans. Knowl. Data Eng.

    (2007)
  • M. Ester et al.

    A density-based algorithm for discovering clusters in large spatial databases with noise

    Proceedings of the 2nd International Conference on Knowledge Discovery and Data Mining

    (1996)
  • E.F. Glynn et al.

    Detecting periodic patterns in unevenly spaced gene expression time series using Lomb-Scargle periodograms

    Bioinformatics

    (2006)
  • K.-Y. Huang et al.

    Mining periodic patterns in sequence data

  • T. Jindal et al.

    Spatiotemporal periodical pattern mining in traffic data

    Proceedings of the 2nd ACM SIGKDD International Workshop on Urban Computing, UrbComp ’13

    (2013)
  • S. Kullback et al.

    On information and sufficiency

    Ann. Math. Stat.

    (1951)
  • J.-G. Lee et al.

    Trajectory clustering: a partition-and-group framework

    Proceedings of the 2007 ACM SIGMOD international conference on Management of data, SIGMOD ’07

    (2007)
  • C. Legány et al.

    Cluster validity measurement techniques

    Proceedings of the 5th WSEAS International Conference on Artificial Intelligence, Knowledge Engineering and Data Bases, AIKED’06

    (2006)
  • Cited by (0)

    View full text