Academic journal article Journal of Electronic Commerce Research

Mining User Movement Similarity Based on Massive GPS Trajectory Data with Temporal Effects

Academic journal article Journal of Electronic Commerce Research

Mining User Movement Similarity Based on Massive GPS Trajectory Data with Temporal Effects

Article excerpt

(ProQuest: ... denotes formulae omitted.)

1.Introduction

As the popularity of smart phones and other GPS-enabled mobile devices continues to grow, location acquisition technologies have become increasingly pervasive, leading to the collection of large spatio-temporal datasets about user movement behavior [Lin et al. 2014; Koutsiouris et al. 2016]. Relatively easy access to large amounts of spatiotemporal data, specifically GPS trajectories, provides an opportunity to discover valuable geographic information concerning individual mobility. In turn, deeper understanding of user movement behavior provides enormous business opportunities with regard to geographic navigation and location-based recommendations [Zheng 2011] by recognizing numerous traffic activities from both the pedestrian side and the transportation side [Liao et al. 2005; Ghourchain 2016].

Recently, user movement similarity has become particularly significant for location-based social network recommendations [Li et al. 2008; Cho et al. 2011] and human mobility prediction [Do et al. 2015], Consequently, research that measures user movement similarity based on travel trajectories has attracted considerable attention [Lv et al. 2013; Chen et al. 2014]. However, for current GPS-enabled applications, collecting information about the similarity of users' trajectories is often difficult and inefficient for the following three reasons:

(1) First, when users turn the GPS-enabled devices on and off casually, the recorded GPS data are often nonuniform, sparse, or lost, and the data collected may be inconsistent with the end-points. Therefore, any two trajectories might not be identical even if they recorded the same path. For instance, see the sample trajectories of TR\ and TR2 in Figure 1. They have been used to record the movements of user1 and user2 on the same path. However, the GPS data points in TR1 and TR2 are rarely identical.

(2) Second, large amounts of GPS point data are recorded in a trajectory (especially in the case of high recording frequency), but only a few of the data are key to exhibiting interesting geographic information about a user's travel [Zheng 2015]. For example, in Figure 1, the "meaningless roaming" data points on TR1 may be motion noise involved in userl's movement.

(3) Third, human geographical movement has always exhibited significant temporal characteristics that are strongly related to the locations [Ye 2011]. For example, the intentions of different users to visit the same place may not be the same, so the time of day of their visits may also vary widely.

To mine the movement similarity of users efficiently, in this paper, first we introduced an efficient GPS trajectory partition method [Yuan et al. 2014] to trim the sequential GPS data into line segments, taking the key end points as the characteristic points for clustering (See Figure 2(a)). Based on these points, we could map all the trajectories onto a series of abstract trip routes. Next, by taking each of the clusters as a fixed territory (location), a user's trajectory reflected his visits to a certain series of geographic locations. Such a "user-visiting-location" relationship may be represented as a user-location matrix (see Figure 2(b)). Finally, taking temporal effects into consideration, we proposed a novel low-rank matrix factorization based method to solve the problem of mining users' similar movements.

The remainder of this paper is organized as follows. Section 2 summarizes significant related work. Section 3 details the novel methods for trajectory partition and fixed territory clustering. Section 4 presents the procedure for mining users' movement similarity with temporal effects. Section 5 shows the experimental results, while Section 6 presents conclusions based on our work.

2.Related Work

2.1. Trajectory information mining

Generally, GPS trajectory data are recorded with very high frequency that provides notably fine-grained information about a movement. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.