1 I NTRODUCTION 1.1 Background
1.4 Defining key terms
1.4.1 Contentious meaning(s) of ‘space’ and ‘place’
The preceding sections have detailed the background to this study (Section 1.1, p1), introduced social media data (Section 1.2, p11) and discussed whether, and how, social media data might offer any geographical value to researchers (Section 1.3, p14). Concepts surrounding the identification of ‘space’ and ‘place’ in social media (meta)data have been outlined, yet these two terms – whose nuanced and
18 multifarious meanings are amongst the most heavily contested in academic
geography and related social science disciplines (Hubbard & Kitchin, 2011) – require further definition as they apply to the current research. Definitions of several key terms used throughout this thesis are given in the following section.
1.4.2 Definition of key terms used in this thesis
Agnew (2011) has suggested that ‘place becomes a particular or lived space’ as humans associate a ‘location somewhere’, or their occupation of that location, with its spatial (or locational) position. Place, therefore, may take on multiple meanings or refer to spaces – the living room, the home, the home town or the country – which have very different geographical extents and which may vary, conceptually, from one person to another. Social media interactions holding discursive text, or text-based metadata, are not immune from such semantic or conceptual
dichotomies; a Twitter tweet containing the word ‘Kansas’ may refer either to the US State of that name or to Kansas City, or to both. Equally, depending upon
context, the mention might be shorthand referring to a popular baseball or football team located in the State or City in question, or to some other location or logical entity altogether (e.g., the Wildcats football team of Kansas State University). Place has social meaning, but determining the context within which a place is mentioned in social media message text, and the exact meaning implicit in that mention of place, is not necessarily a straightforward task.
Senses of known-place(s), affirmed-place(s) and space(s), some of which may be accompanied by apparently ‘accurate’ Latitude and Longitude coordinates, are often highly conflated in social media data. Users of Twitter, for example, when registering, are asked ‘Where in the world are you?’ (Hecht et al., 2011) and may just as reasonably answer ‘BRICK city bitch’ or ‘Somewhere, Overthere’ as ‘Concord, NC’ or ‘iPhone: 40.699490,-73.891556’. Difficulties inherent in identifying and parsing Potential Geographic Information (PGI) in free-form social media message text and associated (meta)data are amplified considerably when, as in this research,
19 place-based geographical references must be detected computationally (Section 4.4, p147). Consequently, and as huge data volumes preclude individual human examination of over ~8 million social media interactions, necessarily focused definitions of ‘space’ and ‘place’ are adopted in this research:
• Space – Refers in this thesis to geographically and explicitly locational data, i.e., to a point (most often) or a geographical extent (much less frequently) defined by one or more pairs of Latitude and Longitude coordinates. Almost all space-based data emanates directly from the small subset of coordinate-geotagged social media interactions in the research data corpus. However, additional spatial data (or ‘spatialities’) may be inferred from
non-coordinate-geotagged interactions either by post-processing (meta)data or by ‘geoparsing’ message text to append coordinates, where possible, to detected references of place.
• Place – Refers in this thesis to computationally-identifiable geographical references in text, i.e., to toponymic place names, e.g., of towns, cities, counties, states or countries etc. Information Extraction (IE) and Named Entity Recognition (NER) techniques from Natural Language Processing (NLP) are used to detect such geographical references in social media data and linked/shared content. The software systems used either rely upon large, open-source gazetteers of toponymic place names (e.g., the ~11 million records available from GeoNames, 2016) or use smaller gazetteers supplemented by logical ‘rules’ (Tear & Maynard, personal communication, 2018) to boost place identification based upon the co-occurrence of certain terms (e.g., ‘Isle of…’, ‘Mount…’, ‘Cape…’) associated with place names.
Space, where it exists in social media interaction (meta)data, may generally be regarded unambiguously; the Latitude and Longitude coordinates of a user’s location have been recorded alongside their message text by a GPS-equipped mobile device just at the moment of message creation. Exceptions exist, of course,
20 such as the production of coordinate-geotagged messages by robotic networks (‘botnets’), described by Marechal (2016) and exemplified by Echeverría & Zhou's (2017) detection of the ‘Star Wars’ botnet, responsible for creating 1.2 million coordinate-geotagged Twitter tweets in North America and Western Europe; or through the presence of unlikely, or nonsensical, spatial coordinates such as the 227 interactions with 0 Latitude and 0 Longitude in the research data corpus.
Place, in social media data, retains many of the elements of ambiguity identified by Tuan (2001) and other geographical theorists but is referenced, on the admittedly narrower grounds adopted here, much more widely in message text, metadata and linked/shared content than space (Section 5.2.2, p190 and 5.2.3, p205). In addition, there is both, a) some overlap between ‘space’ and ‘place’ in digital social media data, and; b) some potential to ‘move’ from space to place, or vice versa. For example:
a) The metadata of some OSN interactions records users’ time zone offsets relative to Greenwich Mean Time (GMT) in seconds. While these values (e.g., 14,400 or -18,000) are neither explicitly space- or place-based,
converting seconds to hours and minutes allows data fusion with world time zone boundaries enabling the small-scale worldwide mapping of online activity (Figure 4-22 and Figure 4-23, p181).
b) The Latitude and Longitude coordinate pairs deposited alongside indvidual’s geotagged interactions may, likewise, be ‘fused’ to official areal units such as US or UK Census boundaries using GIScience techniques (Section 6.4.4, p262). This process enables other types of imputation and reporting which, by exploiting the geographical hierarchy implicit in US and UK Census data (e.g., US Census Tracts aggregate to Counties and States; UK Output Areas to Wards, Local Authorities and Counties etc.) may yield place-based results from purely spatial coordinate data. Conversely, and as an example of the movement from place to space, all toponymic place names detected in
21 message text and successfully geoparsed may be mapped to expand the locational scope of the case study data sets beyond just the small subset of spatially coordinate-geotagged social media interactions present in the research data corpus (Figure 5-12, p224 and Figure 5-13, p224).
A final term, used throughout this thesis, encapsulates the different space- and place-based meanings in social media data outlined above:
• Geographicality – Refers in this thesis to the multiplicity of geographical forms of expression evident in social media data, ranging considerably both, a) in scale, from world time zones covering parts of continents to point-based locations of message creation, and; b) in nature, e.g., incorporating mentions of place(s), again at many different scales, either in message text or linked/shared content, which may or may not be amenable to spatial augmentation through geoparsing. Measuring and scoring geographicality in social media (meta)data (Section 4.6.1, p164) enables cross-comparison of space- and place-based facets of geographical expression at several levels;
by case study event; by OSN platform; by user and, most atomically; by interaction (i.e, individual message and metadata bundle). The results of this work are presented in Chapter 5 (p186) with additional findings presented in Chapter 6 (p227).
In their Introduction to Key Thinkers on Space and Place, a bibliographic
compendium detailing theoretical contributions from 66 scholars of geography and related social science disciplines, Hubbard & Kitchin (2011, p7) state that ‘given the way space and place have been operationalised, they remain relatively diffuse, ill-defined and inchoate concepts.’ In this thesis, meanings of ‘space’ and ‘place’ are measured and operationalised much less diffusely, having been clearly defined above. While necessarily focused, the definitions adopted here enable machine-based classification of very large volumes of social media data; affording an opportunity to determine how space and place are used online, and whether
22 different user groups – especially the most-spatial, coordinate-geotagging, users of two popular Online Social Network platforms – make differential references to place. The rationale for conducting this research is set out below. The relevance of determining who makes the most mention of place in message text, or links to and shares content making the most mention of place via online social media channels, is detailed later in Section 1.6 (p30).