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

This paper presents a preliminary empirical evaluation of the strategic importance of infusing Twitter social media data into classifications of small areas, as a way of moving beyond the nighttime residential geographies of conventional geodemographic classifications. We attempt an empirically based critique of the merits and drawbacks of the use of social media data, in which the value of high spatial and temporal granularity of revealed activity patterns is contrasted with the paucity of individual attribute information. We apply new and novel methods to enrich the profiles of Twitter users in order to generalize about activity patterns in London, our case-study city. More insidious problems in the use of social media data arise from the as-yet-unknown sources and operation of bias in their user bases. Our contribution is to begin to identify and assess the biases inherent in social media usage in social research, and use these to evaluate their deployment in research applications.

This paper explores the age and gender distributions of the bearers of British forenames and identifies key trends in British naming conventions. Age and gender characteristics are known to greatly influence consumption behaviour, and so extracting and using names to indicate these characteristics from consumer datasets is of clear value to the retail and marketing industries. Data representing over 17 million individuals sourced from birth certificates and market data have been modelled to estimate the total age and gender distributions of 32,000 unique forenames in Britain. When aggregated into five year age bands for each gender, the data reveal distinctive age profiles for different names, which are largely a product of the rise and decline in popularity of different baby names over the past 90 years. The names database produced can be used to infer the expected age and gender structures of many consumer datasets, as well as to anticipate key characteristics of consumers at the level of the individual

With as many as 4 million passenger journeys within the London Underground system every weekday, the advertisement spaces across the stations hold considerable potential. However, the planning of specific advertisements across time and space is difficult to optimize as little is known about passers-by. Therefore, in order to generate detailed and quantifiable spatio-temporal information which is particular to each station area, we have explored local social media data. This research demonstrates how local interests can be mined from geotagged Tweets by using Latent Dirichlet Allocation, an unsupervised topic modelling method. The relative popularity of each of the key topics is then explored spatially and temporally between the station areas. Overall, this research demonstrates the value of using Geographical Information System and text-mining techniques to generate valuable spatio-temporal information on popular interests from Twitter data.

There were 30.7 million registered cars in Great Britain in 2011, outnumbering the total number of households recorded by the census. Despite this, the Driving and Vehicle Licensing Agency’s (DVLA) database of car model registrations remains underexplored as an indicator of socio-economic characteristics. In the past, car ownership itself has been frequently considered as a census proxy variable for affluence. However, this is an increasingly dated interpretation as ownership has become more widespread across society and the value of cars varies considerably. Understanding the geography of different car types, however, is likely to be more informative of local population characteristics as the choice of model is dependent on several factors, notably including the cost and the purpose of the vehicle. In partnership with the Department for Transport (DfT), a car segmentation was produced that grouped every car model registered in England and Wales in 2011 into 10 distinctive categories based on the vehicle’s key characteristics. Data representing the total number of registered cars for each car segment and three age groups were made available at a small area geography (known as lower layer super output areas – LSOAs) to be analysed for this study. It revealed that each car segment is uniquely distributed across London, and the rest of England and Wales. The patterns were then compared with key 2011 Census variables on socio-economics to understand the extent to which spatial patterns of broad car characteristics correspond with variances in indicators of social make-up.

A large share of human activity is now routinely captured and stored by commercial organisations as consumer data. These new forms of data have provided many new and exciting opportunities for understanding population trends at fine spatial scales. In many settings, they have reduced our dependence on theory and traditional modelling approaches and fundamentally changed how geographers approach producing representations of the population. However, consumer data are typically created for commercial benefit and do not have academic standard data quality controls, and so, in repurposing the data for social research, we encounter inherent issues of veracity. Without an understanding of uncertainty, social scientists can risk overstretching consumer datasets beyond the specific populations or phenomena they directly pertain to. Moreover, the technical characteristics of large and complex datasets also make it challenging to generate valid information efficiently. Therefore, this paper reviews the major challenges to harnessing consumer data to produce valid spatial representations of the population at large.