Visualizing social media’s impact on local communities

We hear a great deal about ‘big data’ and the ways it might improve our lives. Social media activity including Twitter, blogs and online message boards, produce enormous quantities of data about citizens’ preferences and how they live their lives. Yet, the relationship of social media to local communities and local government remains relatively under explored, as does its potential to illuminate perceptions of place and community. Massive datasets and techniques such as sentiment analysis can produce important insights (for example, Pak and Paroubek, 2010), but what is going on at the human scale?

In this visual essay, we present an aspect of our research where a series of visualizations are used to illustrate different scales of social media activity within south London, UK. Our research has lead to the development of software called localnets.org – a tool that allows interactive exploration and visualization of data. These visualizations are designed to address several challenges in communicating the underlying data. One goal of this research is to assist local policy makers and one way of doing this is through visualizations which have the power to capture attention in a way that tables of data or written descriptions do not. Analysing social media allows us to compare the way people perceive their environment with the way traditional maps delineate it. This is presented using cartographic visualizations. Much of the data derived from social media represents networks, and this essay will also explore the benefits and limits of presenting network graph visualizations.

The Brockley Central blog, alongside websites such as Kentish Towner and Brixton Blog, is a well-established, locally-focused (hyperlocal) London site, making it a good place to begin. ¹ Many existing representations of hyperlocal media sites use points – pins in maps – to visualize where they are located (for example http://hyperlocal.uk/). In Figure 1 we show the sites as spatially extended, covering a particular area. This visualization uses the descriptions the sites themselves give of the area they cover, using for example, ward and postcode boundaries that relate to the areas they describe as their territory.

OPEN IN VIEWER

Figure 1.

Spatial extent of three hyperlocal blogs, using postcode and ward boundary data relating to their self-stated geographical coverage.

A more detailed analysis suggests fuzzier boundaries between different hyperlocal sites. The map in Figure 2 shows location data from four online message boards in South London. A sample of posts from each was manually coded for relevance to a specific geographic location. The ‘heatmap’ style of visualization is often used to show geographic intensity, here it shows the frequency with which post occurs in which location.

OPEN IN VIEWER

Figure 2.

Heatmap of locations referred to in posts on four blogs.

The coverage of these sites is concentrated in local centres, but there are fuzzy, overlapping boundaries with neighbouring areas. We can see that Herne Hill Forum (blue) has an island of activity at the centre of East Dulwich (black), and other similar irregularities can be seen between the other sites. From the map we can start to sense how far local centres extend their influence, and where residents themselves consider their boundaries to be.

This data only reflects the views of those who participate in the sites. However, they do have substantial uptake and may well reflect more widely held perceptions. Herne Hill Forum, for example, estimate they reach 50 per cent of the adult population of Herne Hill through their mailing list, website and Twitter. When a media presence reaches this level, it may even go from passively reflecting perceptions of boundaries and begin to start actively informing them. We also wanted to look at how this distribution of coverage might relate to the built environment. Locality is often conceptualized as a radius around a central point like a train station. Significant evidence suggests that the way people perceive closeness in cities is actually more to do with the complexity of a route between point and another (see Turner, 2007).

This map (Figure 3) uses an angular analysis of the street network to represent routes that are more accessible from Brockley station in darker tones (dark grey streets), even if they are further away, and show streets that are expected to be perceived as further away in a lighter grey. This approach more closely approximates psychological conceptions of distance than a metric radius ‘as the crow flies’. Overlaid is a ‘heatmap’ of posts on the Brockley central map, using the same method as described above. The main road stretching to the south is psychologically and spatially better linked with the centre of Brockley, as indicated by the darker grey roads. We can see that the blog also tends to have posts that follow the road, which forms the social and spatial focal linking a number of different neighbourhoods all of which are well covered by Brockley Central. To the east of Brockley station, on the other hand, lies part of Brockley that is segregated by train lines running north–south and east–west. Here, the blog has little to report. The virtual might not, after all, be so able to transcend spatial divides as is often thought, and here actually serves to reinforce barriers in the built environment.

OPEN IN VIEWER

Figure 3.

Heatmap of locations referred to in blog posts on Brockley Central throughout 2013–2014 overlaid with angular analysis of street network centred on Brockley station.

Message boards and blogs only offer a certain level of granularity. We can use Twitter behaviour to see in more detail how communication networks are formed in Brockley. The Brockley Central blog also has a Twitter account. The cloud-like network graph (Figure 5) shows all of Brockley Central’s Twitter followers. The size of the discs shows how many followers each follower account has within that network, and the links between them show where one Twitter user follows another. Colours indicate communities within the network, which were identified algorithmically purely through analysis of the density of connections on Twitter.

This network graph is almost illegible,and the colours seem almost arbitrary. Network diagrams – even ones as confused as this – have the power to capture the attention, for example, when articulating this work to policy makers. Possibly all they convey is the quantity of data that is being processed. The map (Figure 4) shows an improved way of representing this data. In this case, a sample of the top 5 per cent most connected profiles mapped their location on the map wherever possible. The colours for each node are the same as those in the ‘cloud’ network diagram (Figure 5). Most private individuals do not reveal their location on Twitter as located profiles are mainly local businesses. ²

OPEN IN VIEWER

Figure 4.

Map of location-specific profiles following @brockleycentral. Size shows number of connections. Colour shows network community within @brockleycentral’s Twitter followers.

OPEN IN VIEWER

Figure 5.

Network ‘cloud’ showing connections between Brockley Central’s Twitter followers.

Filtering and spatializing the data makes it easier to make sense of. Network concentrations turned out also to be spatial concentrations, suggesting that even within the wider community of Brockley Central Twitter followers there are tight sub-communities of nearby businesses that follow one another more than they do those that are further away. This may seem commonsense; however, communication technology has often been conceived as making distance irrelevant so it is perhaps surprising to see it being used to create tight networks that are even more proximate than the hyperlocal (see Cairncross, 2001).

In a separate analysis, we can see who interacts with whom on Twitter using replies and retweets. Twitter users were discovered by using Brockley Central as a ‘seed’ and then ‘snowballing’ by looking for Twitter users who interact with Brockley Central, who then interacts with accounts that interact with that account and, so on. Tracing these connections over time, we begin to see key brokers of local online interaction emerge, many of which are local cafes and restaurants like Jam Circus, Masala Wala and The Orchard (Figures 6 and 7). Their position as visual landmarks and shared spaces on the high street translates into a role as generators of mediated interactions.

OPEN IN VIEWER

Figure 6.

Twitter interaction network, seeded with the Brockley Central account. Colours show algorithmically identified ‘communities’. Thickness of lines shows number of interactions. Created using localnets.org.

OPEN IN VIEWER

Figure 7.

Twitter interaction network around Brockley Central, focusing on JaneCanDoSE4. Created using localnets.org.

On the Twitter network graph we can see JaneCanDoSE4 (SE4 is the local postcode). She runs Crotfon Park blog, which covers a much smaller spatial scale than Brockley Central. As a result, she is able to keep a strong presence both online and in person. In interview, she described how she constantly gleans information by physically visiting businesses and developing strong face-to-face relationships with their owners who share information. ³ She then passes this on via her blog but also in conversation, becoming an essential node in both online and offline communication networks, and interfacing between these realms. As blogging becomes more local, its relationship with everyday space becomes closer. Unlike Brockley Central – which focuses on receiving and distributing information online – Jane establishes herself as a personality physically and virtually,using Crofton Park’s small parade of shops as a geographic centre. These more fine-grained network visualizations contrast with the ‘cloud’ of connections shown above. Network graphs such as these become more legible when they have fewer nodes, and can start to convey useful information beyond articulating the sheer quantity of data being processed.

Interviews revealed two streets in Brockley that had email lists (Figure 8) for sharing recommendations, news and discussion. One interviewee described how a particularly proactive neighbour sustained a list for her street by knocking on the door of new arrivals and inviting them to sign up. At this scale, connections are established in person before technology comes into play. Unlike the loose associations of Brockley Central, where individual residents come into contact with stories online but not necessarily one another, communication here is direct and interpersonal, strengthening neighbourly cooperation. Figure 8 suggests how communication settings at different spatial scales nest within one another. Streets with mailing lists benefit from falling within the wider coverage of Brockley Central. They circulate information in the form of links to the blog which, if they impinge upon the street, become the stimulus for interaction and even coordination of face-to-face meetings. However, this closer community can introduce new social pressures.In our study, we heard from one interviewee that new residents refusing to join the list were perceived as un-neighbourly. The existence of a means to communicate can clearly translate into an expectation to do so, that works against the ‘metropolitan mindset’ of detachment and anonymity that, according to Simmel (1903), is one of the fundamental social characteristics of the modern city.

OPEN IN VIEWER

Figure 8.

Spatial extents of different mediated communication settings.

There is a great deal of civic potential in these new forms of communication. This brief research overview is intended to show that interactions and behaviours that were once invisible or did not exist are now available for analysis. When visualized, this data can help illuminate communication patterns in a neighbourhood and their relationship to the built environment resulting in a potential tool for architects, planners, developers and local authorities.

Visualization can also help local policy makers understand what issues are prevalent locally and who is discussing them, helping to provide services that are more responsive to changes in citizens’ priorities.

The quantity of data derived from hyperlocal social media is so large that new ways of representing and navigating it are necessary. It would be prohibitively resource intensive for anyone interested in a particular area to read every tweet or blog about it. Network diagrams and ‘heatmaps’ are two ways that a corpus of tweets can be aggregated to show the bigger picture. They can also direct attention to important material. For example, if the intensity of activity around a street greatly increases, it can be flagged for further analysis.

The tools which citizens use to discuss local issues – Facebook, Twitter, blogs – and the tools that are used to aggregate and visualize those discussions have clear implications for local democracy. For example, Facebook is not included in this analysis because, even when discussions on the site are notionally public, retrieving the data is prohibitively complex. By clearly illustrating the scale and value of hyperlocal discourse, it is hoped that discussions of the platforms on which it takes place can be better informed. Looking to the future, it seems highly likely that social media will only grow, even if its exact forms are unpredictable. Clear visualizations of the underlying data will be vital to informed discussion of this phenomenon.