A multi-scale approach to exploring urban places in geotagged photographs
Introduction
User-generated content (UGC), information or media that citizens use, create, and share online, has evolved from what many viewed as a curiosity or a passing technology-led fad in the early days of the Web 2.0 era to now being recognized as an important element of many government, business, scientific and social processes (Elwood et al., 2012, Hermida, 2010, McKenzie et al., 2012, Shirkey, 2008, Tapscott and Williams, 2006). Increasingly, the production and communication of information and knowledge are becoming more collaborative and rooted in networked communities as the distinctions between data users and data producers and, by extension, experts and amateurs are blurred. A small, but growing proportion of UGC contains geographic information either in the form of explicit spatial coordinates generated from personal locational devices (e.g. mobile phones, GPS units, etc.) or less explicit references such as names of landmarks, regions or cities.
Some forms of spatial UGC, such as users’ mapping of animal sightings, drivers’ reports of potholes in roads, and citizens’ comments on land management issues, have direct linkages to specific data products, community interests or “citizen science” initiatives (Connors et al., 2011, Wiersma, 2010). This spatial subset of UGC is often referred to within the GIS and GIScience literature as Volunteered Geographic Information (VGI) community following Goodchild (2007). The linkages are less evident with other forms of spatial UGC (e.g. photos, videos and Twitter posts with geographic references) that are typically created simply as an outflow of personal interests or web-based communication and may not be not viewed by their authors as geographic data in their own right (Feick and Roche, 2012, Purves et al., 2011). With much of the social web now enabled with location sensors, the social web is fast becoming the social geo-web (Sui & Goodchild, 2011), and geographers and others have expressed great interest in the opportunities and dangers associated with these vast new sources of data (e.g. Wilson, Gosling, & Graham, 2012).
A prime example of the spatialization of social-web interactions is the proliferation of production and access to digital photographs encoded with geographic information and the subsequent exploitation of these repositories for exploratory, national and global-level research (Crandall et al., 2009, Zhang et al., 2012). Much of this research has centerd on exploring what types of information can be derived from the growing volumes of geotagged photographs (GTP) that are uploaded and shared on sites such as Panoramio, Instagram and Flickr (Antoniou et al., 2010, Kennedy et al., 2007). GTPs are comprised of a georeferenced image and a set of descriptive keywords or tags that users add to describe the photo, the place or event it is associated with, or its personal meaning. Since users determine what they photograph, what tags they use, and which photos they share, GTPs hold promise as a rich data source for investigating how people perceive and characterize the phenomena they photograph and, potentially, for uncovering hidden geographies and spatial structures in social processes (Ferarri, Rosi, Mamei, & Zambonelli, 2011).
In this paper, we propose a multi-scale method for exploring patterns in the spatial and thematic content of GTPs. Our intent is exploratory in nature and is directed at distilling new insights or hypotheses concerning localized, or scale-dependent, expressions of place that are encoded within individuals’ GTPs. We build upon a sizeable body of research centered on GTPs including recent work that investigate: concepts of place and vernacular geographies across space and time in Flickr and Geograph (geograph.org.uk/) photographs (Dykes et al., 2008, Hollenstein and Purves, 2010, Purves et al., 2011), relationships between physical and cyber spaces (Graham & Zook, 2011), individuals’ patterns of movement in urban environments (Andrienko and Andrienko, 2011, Jankowski et al., 2010, Kisilevich et al., 2010) and how tagged social media can be mined to discern non-experts’ approaches to classifying images (Rorissa, 2010) or social trends (Jin, Gallagher, Cao, Luo, & Han, 2010), among others.
A significant thread in the research on GTPs has focused on developing and evaluating measures of tag similarity in order to characterize massive numbers of GTPs (Crandall and Snavely, 2012, Crandall et al., 2009) or to support online tools such as tag recommendation engines, query tools, and visualizations of tag semantics (Moxley et al., 2008, Wu et al., 2008). Indeed, much of the research to date using GTPs has been computational in nature, rather than analytically focused, although some recent examples show considerable promise (e.g. De Choudhury et al., 2010). We suggest that the potential exists to generate a finer understanding of local environments through investigation of the spatial, temporal, and semantic qualities of GTPs. Rattenbury and Naaman (2009), for example, demonstrate an innovative approach for distilling place semantics from GTP collections using tag- and spatial-scanning approaches that identify locations where significant concentrations or “bursts” of place-related tags are found.
We approach this dimension of localized characterization of space and place by applying a metric adapted from classical odds ratios, used in ecology and epidemiological studies to measure exposure effect, to search for “tag-space” neighborhoods within an urban environment. We are interested here in exploring whether an ecological approach can help quantify the intrinsic spatial and thematic properties of GTPs and uncover areas of tag similarity. In contrast to much of the other research of this type, we use a multi-resolution approach to GTP aggregation that, we suggest, facilitates exploration of urban patterns and processes that occur across different geographies or “scales” of analysis (e.g. streetscapes, neighborhoods, regions). We believe that this approach has potential to help us to understand, to some degree, the strength and spatial extents of citizens’ perceptions, filtering and cognition of urban processes and forms. In this way, we may be able to situate previous GTP research related to spatializing place via space–time patterns in Flickr photos (e.g. Crandall et al., 2009, Kennedy et al., 2007) within a broader context of small scale patterns associated with GTPs. Additionally, large scale studies such as those tracing individuals’ movement through urban space from temporal trajectories of GTPs (e.g. De Choudhury et al., 2010, Jankowski et al., 2010) may fit another class of scale-specific processes associated with certain landscapes.
In the following pages, we demonstrate our approach to GTP analysis using data obtained for Vancouver, Canada from the Flickr API. Prior to discussing the methods used, we first describe the study area selected for analysis and provide some context related to spatial and temporal changes to the study area during the study period. We then describe our data processing methods and the building of a final GTP database in detail. Following a description of the data analysis methodology, we provide a series of results examining trends in tagging of GTP across space and using tessellations of varying resolutions. We conclude the paper with a discussion of our key findings and highlight caveats of the current analysis and areas for further research.
Section snippets
Methods
In this analysis, we are interested in the joint analysis of space, scale and meaning embedded in GTPs obtained for our study area. Scale is handled in a straightforward manner, by making observations at numerous spatial aggregations and comparing how measures change across scales, as is common in biogeographical and ecological research (e.g. Fortin and Dale, 2005, Turner et al., 2001). Note that in this context, the term “scale” is used as an indication of the resolution at which space is
Results and discussion
As expected, the spatial distribution of GTPs is clustered in the main urban tourist and entertainment oriented core of the city (Fig. 3). As noted earlier, the total number of photos obtained for the study was 54,522. Within 1 km of a central downtown intersection (i.e. Burrard St & Robson St.) in downtown Vancouver, some 12,000 photos (22%) were found, while 41,054 photos (75%) were found within 5 km. The maps in Fig. 3 depict geometric five-class symbology, which is a classification scheme
Conclusion
This paper demonstrates a method for analyzing tag frequency for a set of GTPs across multiple scales using Flickr data obtained for the city of Vancouver, Canada. In general terms, we demonstrated the dependence of tag characteristics on spatial scale of aggregation. At larger areas of aggregation, tag-space is dominated by a few frequent tags that describe large geographies, whereas more place-specific tags emerged at local scales. In the context of the Vancouver Flickr data, this effect was
Acknowledgements
The authors would like to thank the Social Sciences and Humanities Research Council (SSHRC) of Canada for supporting this research. In addition, the authors are grateful to the three anonymous reviewers for insightful suggestions that have substantially improved this article.
References (57)
Spaces of transition spaces of tomorrow: Making a sustainable future in Southeast False Creek, Vancouver
Cities
(2007)- et al.
Multiscale assessment of landscapes and watersheds with synoptic multivariate spatial data in environmental and ecological statistics
Mathematical and Computer Modelling
(2000) - et al.
Spatial generalization and aggregation of massive movement data
IEEE Transactions on Visualization and Computer Graphics
(2011) - et al.
Web 2.0 geotagged photos: Assessing the spatial dimension of the phenomenon
Geomatica
(2010) Living first in downtown Vancouver
Zoning News
(2000)- Black, J. (2009, October 16). A two-year march against homelessness. The Tyee. Retrieved from...
- et al.
Global and local spatial autocorrelation in bounded regular tessellations
Journal of Geographical Systems
(2000) - CBC News (2012, December 26). “Vancouver’s Downtown Eastside feeling gentrification squeeze” - British Columbia - CBC...
- et al.
Open source databases and their spatial extensions
- et al.
Volunteered geographic information: The nature and motivation of produsers
International Journal of Spatial Data Infrastructure Research
(2009)
Citizen science in the age of neogeography: Utilizing volunteered geographic information for environmental monitoring
Annals of the Association of American Geographers
Modeling people and places with internet photo collections
Communications of the ACM
Exploring volunteered geographic information to describe place: Visualization of the ‘Geograph British Isles’, GISRUK 2008
Researching volunteered geographic information: Spatial data, geographic research and new social practice
Annals of the Association of American Geographers
Understanding the value of VGI
Spatial analysis: A guide for ecologists
A scoping review of spatial cluster analysis techniques for point-event data
Geospatial Health
Quality assessment of the French OpenStreetMap dataset
Transactions in GIS
Citizens as sensors: The world of volunteered geography
GeoJournal
Visualizing global cyberscapes: Mapping user-generated placemarks
Journal of Urban Technology
How good is volunteered geographic information? A comparative study of OpenStreetMap and Ordnance Survey datasets
Environment and Planning B
Twittering the news
Journalism Practice
Exploring place through user-generated content: Using Flickr tags to describe city cores
Journal of Spatial Information Science
Cited by (60)
Streets of London: Using Flickr and OpenStreetMap to build an interactive image of the city
2020, Computers, Environment and Urban SystemsCitation Excerpt :Straumann, Cöltekin, and Andrienko (2014) use temporal and user information to build trajectories and compare group behaviours and thus, they argue, explore narratives in the city. Feick and Robertson (2015) make two important observations—firstly, the semantics derived from georeferenced image tags is dependent on the scale of the analysis unit and, secondly, the distribution of images is strongly influenced by the street network (and open spaces). This second observation makes it all the more surprising in our view that most studies to date have linked urban properties derived from UGC with space ignoring the underlying street network.
Exploring spatio-temporal changes of city inbound tourism flow: The case of Shanghai, China
2020, Tourism Management