An invited talk by Dr Rene Westerholt at the Quantitative Spatial Science Group (School of Geographical Sciences, University of Bristol) next Monday, 29 April.
Spatial Analysis of Geosocial Media Data–Methodological Remarks
The profound digital transformation of our lives has led to the creation of new datasets that together form a kind of "digital skin of our earth". One of these types of datasets that have emerged recently is geosocial media, i.e. social media posts that contain geographic content and are annotated by GPS coordinates. Geographers and cognate scholars have at their disposal a rich, but technically sophisticated source of information that partly reflects the everyday life of ordinary people. In this talk, methodological aspects will be discussed that are relevant for the spatial analysis of geosocial media data. The focus will thereby be on spatial autocorrelation. This characteristic is fundamental for geographical structure and pattern recognition and thus ensures a broad relevance of the results presented. The talk discusses how the spatial characteristics of geosocial media contributions and the limitations of the information contained in them influence the underlying assumptions of conventional analytical frameworks. It is demonstrated how this infringement of assumptions leads to the disclosure of false patterns and misleading assessments of the strength and nature of spatial associations. The discussion also includes scale issues, which are of particular importance in the geographic context. All these results have been obtained based on two different Twitter datasets, as well as using idealised synthetic data allowing to control spatial characteristics. The talk will conclude with an outlook on possible future directions of spatial analysis (with regard to the analysis of user-generated information), and with establishing a link to the related field of place-based analysis.