The geography of Twitter topics in London

Social media data are increasingly perceived as alternative sources to public attitude surveys because of the volume of available data that are time-stamped and (sometimes) precisely located. Such data can be mined to provide planners, marketers and researchers with useful information about activities and opinions across time and space. However, in their raw form, textual data are still difficult to analyse coherently and Twitter streams pose particular interpretive challenges because they are restricted to just 140 characters. This paper explores the use of an unsupervised learning algorithm to classify geo-tagged Tweets from Inner London recorded during typical weekdays throughout 2013 into a small number of groups, following extensive text cleaning techniques. Our classification identifies 20 distinctive and interpretive topic groupings, which represent key types of Tweets, from describing activities or informal conversations between users, to the use of check-in applets. Our motivation is to use the classification to demonstrate how the nature of the content posted on Twitter varies according to the characteristics of places and users. Topics and attitudes expressed through Tweets are found to vary substantially across Inner London, and by time of day. Some observed variations in behaviour on Twitter can be attributed to the inferred demographic and socio-economic characteristics of users, but place and local activities can also exert a considerable influence. Overall, the classification was found to provide a valuable framework for investigating the content and coverage of Twitter usage across Inner London.