Article Preview
TopIntroduction
The classification of spatial regions such as buildings and districts in an urban setting is essential for urban planning (Xing et al., 2018). Remote sensing techniques have been traditionally considered vital classification methods because of their ability to capture the physical characteristics of spatial regions (Hu, & Wang, 2013; Huang et al., 2014; Huang et al., 2015; Wu et al., 2009; Zhang & Du 2015; Wen et al., 2015). However, because they only focus on the physical characteristics of spatial regions, remote sensing techniques cannot adequately determine the functions of spatial regions (Pei et al., 2014).
Along with the increasing popularity of positioning techniques and their application in social media, there has arisen an abundance of trajectory-related data with location information. These trajectories have a strong correlation to the functions of related spatial regions, which include people’s travel, entertainment, shopping, and other daily habits (Giannotti et al., 2016). Based on the local spatiotemporal characteristics of trajectories, many methods have been developed for classifying the functions of spatial regions.
Pei et al. (2014) and Toole et al. (2012) proposed using mobile phone data to derive information pertaining to land use in spatial regions. Yuan et al. (2012) and Ge et al. (2019) adapted human mobility patterns mined from GPS trajectory datasets to discover regions of different functions in a city. In terms of using public transportation card data, Zhong et al. (2014) made use of the spatial relationships between public transportation trips, stops, and buildings, to establish a two-step framework to infer the functions of buildings, while Zhou et al. (2017) proposed to infer functions occurring around metro subway stations according to staying activities patterns (such as with the same start time, end time, and frequency) derived from smartcard data. More recently, spatially referenced social media data (i.e., Facebook, Twitter, Tencent’s WeChat, and Sina Weibo check-in data) have emerged as new data sources for the classification of spatial regions (Chen et al., 2017; Wu et al., 2018; Gao et al., 2017). Furthermore, classification methods based on multiple data sources (Liu et al., 2018; Tang et al., 2019; Xia & Li 2019; Liu et al., 2017; Zhang et al., 2017) have emerged, as methods such as the above are subject to the coarse spatial resolution data and limited information of land-use types.
However, because methods such as the ones above use only local spatiotemporal characteristics of trajectories located in and/or near the spatial regions, they may generate erroneous results in some cases, because they fail to consider the global connections between the spatial regions. For example, a large international conference center might be mistakenly classified as a shopping mall. The reason for this is that although the trajectories located in these two types of buildings basically have the same local spatiotemporal characteristics, the outside, or farther, trajectories might in fact be significantly different. In particular, the origins of the trajectories involved in shopping malls are more likely to be from the same city, while for international conference centers, the origins might be from other cities or even other countries.