Extracting and understanding urban areas of interest using geotagged photos

https://doi.org/10.1016/j.compenvurbsys.2015.09.001Get rights and content

Highlights

  • We propose a framework for extracting and understanding urban AOI from geotagged photos.

  • We design an experiment to construct optimal polygons from point clusters.

  • We mine knowledge from the extracted AOI, and investigate their spatiotemporal dynamics.

  • An online system has been developed as a proof-of-concept to show the AOI in different cities.

Abstract

Urban areas of interest (AOI) refer to the regions within an urban environment that attract people's attention. Such areas often have high exposure to the general public, and receive a large number of visits. As a result, urban AOI can reveal useful information for city planners, transportation analysts, and location-based service providers to plan new business, extend existing infrastructure, and so forth. Urban AOI exist in people's perception and are defined by behaviors. However, such perception was rarely captured until the Social Web information technology revolution. Social media data record the interactions between users and their surrounding environment, and thus have the potential to uncover interesting urban areas and their underlying spatiotemporal dynamics. This paper presents a coherent framework for extracting and understanding urban AOI based on geotagged photos. Six different cities from six different countries have been selected for this study, and Flickr photo data covering these cities in the past ten years (2004–2014) have been retrieved. We identify AOI using DBSCAN clustering algorithm, understand AOI by extracting distinctive textual tags and preferable photos, and discuss the spatiotemporal dynamics as well as some insights derived from the AOI. An interactive prototype has also been implemented as a proof-of-concept. While Flickr data have been used in this study, the presented framework can also be applied to other geotagged photos.

Introduction

Urban areas of interest (AOI) refer to the areas within an urban environment which attract people's attention. Such areas may contain city landmarks, commercial centers, and recreational zones, or may simply provide a scenic view of the city. The concept of urban AOI is different from urbanized area, as the former puts additional emphasis on people's interests. Consequently, an urbanized area (e.g., a regular residential neighborhood) may not necessarily also be an AOI. Unlike the well-defined administrative districts, the boundaries of urban AOI are vague. This is because urban AOI are subjective: given a familiar city, most people will have a list of interesting areas in mind; yet, their lists may differ due to different ages, cultures, education backgrounds, personal interests, and so forth. Similarly, agreeing on certain AOI does not imply agreeing on their spatial extents and delineations. In this respect, AOI are related to the concept of vague place (Cohn and Gotts, 1996, Jones et al., 2008, Liu et al., 2010, Montello et al., 2003).

Urban AOI have great meanings in multiple application domains. For tourists, AOI highlight the interesting zones within a city, and can therefore be used to support trip planning of travelers. For city planners, AOI reveal the regions which receive high exposure among the general public. Accordingly, these regions could be assigned higher priorities when there are only limited resources for urban planning projects, such as city beautification (Espuche et al., 1991, Gandy, 2006). Since AOI are often visited by a large number of people, transportation analysts can examine these regions to understand traffic flows and human mobility patterns (Batty, 2007, Yuan and Raubal, 2012). In addition, information service providers can display targeted information based on AOI (e.g., highlighting the hotels within the AOI of a city).

AOI exist in the perception of people, and as a result, it is difficult to capture AOI using traditional data collection methods. In urban studies, remote sensing data have been often used to monitor the status of a city (Hsu et al., 2013, Shi et al., 2014, Yu et al., 2010). While good at detecting physical phenomena (e.g., land use types), remote sensing data are unable to observe interests of people. Alternatively, human participant survey, such as the one employed by Montello et al. (2003), could be used to uncover AOI. While such surveys provide valuable insights, they are labor-intensive, time-consuming, and do not scale well.

Social media, such as Twitter and Flickr, record the interaction between people and their surrounding environment (Mckenzie, Adams, & Janowicz, 2013). Compared to remote sensing images, social media data contain valuable information about the behavior of people in geographic space. In most cases, data from social media platforms can be retrieved through their public APIs at low costs. While social media data have often been criticized for the representative issue, i.e., the users may not constitute a representative sample of the entire population (Chou et al., 2009), they are nevertheless generated by millions of people from different countries throughout the world.

Among the many types of social media data, geotagged Flickr photos possess a high suitability for exploring urban AOI. One major advantage of Flickr data is that they reflect the interest of people towards locations (Crandall et al., 2009, Kennedy et al., 2007, Li and Goodchild, 2012). This can be distinguished from some other social media data, such as geotagged tweets, which are not necessarily related to the locations they originate from. For example, a user may tweet about a humanitarian crisis, such as a particular drought, in East Africa from her office in Santa Barbara, California. While the location of her office is attached to that tweet, her real interest is in the events in Africa. In addition, the openness of the Flickr API allows the retrieval of publicly available data throughout the world and since the year of 2004. This adds an interesting temporal component to the data. Many other social media data are either constrained by user permissions (e.g., Facebook) or limited API accessibility for long-term data (e.g., Twitter and Foursquare). Besides, existing research shows that a major proportion of Flickr photos were taken in urban areas (Crandall et al., 2009, Hollenstein and Purves, 2010), and this gives Flickr data one more advantage for studying urban AOI. Back in March 2013, Flickr already had 87 million registered members and more than 8 billion photos (Jeffries, 2013).

This paper aims at extracting urban AOI and understanding them from spatial, temporal, and thematic perspectives. Geotagged Flickr data from six different cities in six different countries have been retrieved for this study. From a spatial perspective, this research examines the locations and extents of AOI in multiple cities and countries. From a temporal perspective, this research investigates the evolution of AOI in the past ten years (2004–2014), and compares the evolution patterns in developed and developing countries. From a thematic perspective, this research extracts the semantics of AOI from the textual tags and user contributed photos, and uncovers the thematic topics underpinning these AOI. The contributions are listed as follows:

  • We develop a framework for extracting and understanding urban AOI from geotagged photo data. This framework is not restricted to Flickr data, but can also be applied to other geotagged photos.

  • To generate proper polygon representations for AOI from point clusters, we design an experiment to identify an optimal parameter for the chi-shape algorithm, which balances the emptiness and the complexity of the generated polygons.

  • We examine the extracted AOI from spatial, temporal, and thematic perspectives. Additional insights, such as the changes of landmark exterior and untypical AOI, are also discovered and discussed.

  • We design and implement an online prototype that allows readers to explore the extracted AOI. This prototype can be accessed at: http://stko-exp.geog.ucsb.edu/urbanAOIs/index.html.

The remainder of this paper is organized as follows. Section 2 discusses related work on points of interest, vague places, volunteered geographic information, as well as extracting hot zones and landmarks from spatial footprints. Section 3 describes the dataset used in this work, and Section 4 presents the framework for extracting and understanding urban AOI from geotagged photos. Section 5 provides a discussion on the spatiotemporal dynamics of AOI and the insights acquired from the extracted AOI. Section 6 describes the interactive prototype implemented based on the proposed framework. Finally, Section 7 summarizes this work and discusses future directions.

Section snippets

AOI, POI, and vague place

The concept of AOI is closely related to two notions in literature, namely point of interest (POI) and vague place. POI represents individual locations (e.g., a restaurant or a landmark) which are of interest to people (Mckenzie et al., 2014, Yoshida et al., 2010). In contrast, an AOI may contain multiple co-located geographic features, such as the restaurants on a pedestrian street or several nearby landmarks (Elias, 2003, Raubal and Winter, 2002). AOI may also include the areas that do not

Dataset

Data used in this study have been retrieved using Flickr's public API (https://www.flickr.com/services/api/). Six cities from six different countries have been selected for this work, which are: New York City (NYC), London, Paris, Shanghai, Mumbai, and Dubai. The first three cities are generally recognized as well-developed cities, while the second three cities have experienced fast development in the past 10 years. We deliberately choose these two groups in order to examine the difference in

Framework overview

A three-layer framework (Fig. 1) has been designed for extracting AOI from geotagged photos and understanding their spatiotemporal dynamics. Learning from the DIKW (data, information, knowledge, and wisdom) Pyramid (Rowley, 2007), the proposed framework aims at deriving knowledge from data following a bottom-up paradigm. The data layer at the bottom is responsible for retrieving data from Flickr's public API and pre-processing the data for later stages. The spatiotemporal layer, lying in the

Results and discussion

Applying the framework to the retrieved Flickr data, we have extracted AOI for the six cities in the past ten years. In this section, we examine these results, and present some insights that can be derived from the AOI. We also discuss the importance and potential applications of the derived insights.

Prototype

While we have discussed some AOI, there are many other interesting areas in our result. Instead of describing them one-by-one, we develop an online prototype (http://stko-exp.geog.ucsb.edu/urbanAOIs/index.html), and invite the readers to interactively explore the AOI and examine the discussed insights. Thus, this prototype has been designed for two purposes: 1) it serves as a supplementary material which provides additional content for the current paper; and 2) it acts as a proof-of-concept for

Conclusions and future work

In this study, we define urban AOI as the areas within a city that attract people's attention. A variety of reasons can contribute to the formation of AOI, such as prominent landmarks, commercial zones, and scenic views. The concept of AOI can be applied to multiple domains, including urban planning and location-based services, and can also be used as an additional layer in a GIS to support spatial queries. This study extracts AOI from geotagged photos and seeks a better understanding of how

Acknowledgment

The authors would like to thank the three anonymous reviewers for their constructive comments and feedbacks.

References (60)

  • T. Brinkhoff et al.

    Measuring the complexity of polygonal objects

  • W.C. Chen et al.

    Visual summaries of popular landmarks from community photo collections

  • Z. Cheng et al.

    Exploring millions of footprints in location sharing services

  • W.Y.S. Chou et al.

    Social media use in the United States: Implications for health communication

    Journal of Medical Internet Research

    (2009)
  • A.G. Cohn et al.

    The ‘egg-yolk’ representation of regions with indeterminate boundaries

    Geographic Objects with Indeterminate Boundaries

    (1996)
  • D.J. Crandall et al.

    Mapping the world's photos

  • D.J. Crandall et al.

    The livehoods project: Utilizing social media to understand the dynamics of a city

  • C. Davies et al.

    User needs and implications for modelling vague named places

    Spatial Cognition & Computation

    (2009)
  • B. Elias

    Extracting landmarks with data mining methods

  • S. Elwood

    Volunteered geographic information: Future research directions motivated by critical, participatory, and feminist GIS

    GeoJournal

    (2008)
  • A.G. Espuche et al.

    Modernization and urban beautification: The 1888 Barcelona World's Fair*

    Planning Perspective

    (1991)
  • M. Ester et al.

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

  • P.F. Fisher

    Extending the applicability of viewsheds in landscape planning

    Photogrammetric Engineering and Remote Sensing

    (1996)
  • P. Frontiera et al.

    A comparison of geometric approaches to assessing spatial similarity for GIR

    International Journal of Geographical Information Science

    (2008)
  • M. Gandy

    4 urban nature and the ecological imaginary

  • S. Gao et al.

    Constructing gazetteers from volunteered big geo-data based on Hadoop

    Computers, Environment and Urban Systems

    (2014)
  • M.F. Goodchild

    Citizens as sensors: The world of volunteered geography

    GeoJournal

    (2007)
  • M.F. Goodchild

    Prospects for a space–time GIS

    Annals of the Association of American Geographers

    (2013)
  • M. Haklay

    How good is volunteered geographical information? A comparative study of OpenStreetMap and Ordnance Survey datasets

    Environment and Planning. B, Planning & Design

    (2010)
  • M. Haklay et al.

    Openstreetmap: User-generated street maps

    Pervasive Computing, IEEE

    (2008)
  • Cited by (0)

    View full text