Scale Model
Find, Follow and Reach Communities on Twitter
![[in beta]](https://miro.medium.com/v2/resize:fill:48:48/1*v-NULuaXus0LWWmACqFrNA.jpeg){kind=small}
In a media ecosystem dominated by networks, gaining an understanding of your audience may be confusing. Who are all those followers, and how can you actually reach specific groups of users? Industry accepted demographic buckets just don’t cut it anymore, especially if the community that you’re trying to reach is much more nuanced than a set of constant variables such as age, income or zip code.
Rather than consider binary group membership — Pepsi drinker or not, U2 fan or not, animal lover or not — it is important to identify someone as positioned within a network of connections and interests, where connections between users and interests vary in strength. Leveraging a networked-informed understanding of users gives us significantly better ability to group them into cohesive segments, as well as discover properties we would have never thought to look for.
Enter Scale Model.
Scale Model helps find, follow and reach networked communities on Twitter. The interface (and underlying APIs) surface the most important content shared within the observed communities and sub-communities in realtime. By consuming twitter data directly through Gnip (data provider recently acquired by Twitter) and scoring content using various graph centrality measures, the interface surfaces relevant and sometimes surprising results.
Our core assumption: important topics bubble up within communities
Given a small set of initial inputs, we construct a model of an online community using graphs representing users and their relationships. Analyzing the shape of these graphs give us the ability to identify meaningful sub-communities, or regions that display significantly high levels of connectivity. The underlying structure of the graph helps us understand the observed community — which group is dominant, which sub-groups are closer together, and which users holds a strategic position.
Network graphs also give us the ability to gauge authority within a given context. Even though Barack Obama is highly influential in the broader context of US politics, he may not be as relevant or important within more specific contexts, such as the Republican party or amongst Ted Cruz supporters. Such scores help both surface the most important content within the dashboard, as well as inform realtime ad targeting for active campaigns.
There’s a beautifully designed dashboard which affords multiple views on the data across the community or within the sub-communities. For example, the big board view (below) shows a tree-map visualization of images shared within the Animal Lovers community (and yes — there most certainly is a community of Turtle lovers out there!). The larger the image, the more important it is within the observed community.
NOT Your Typical Listening Tool.
Here’s how it works.
We begin with a small set of inputs that represent the type of communities we’re trying to reach. If we’re building a community model of the upcoming US 2016 elections, we could use a few Twitter handles — @hillaryclinton @sensanders @randpaul — as well as some popular conversational hashtags — #ReadyForHillary, #Bernie2016 and #CruzCrew. While the initial inputs potentially bias the process, we utilize an iterative approach that takes this into account.
Next we construct a model of the identified community, expanding on the initial set of inputs by leveraging network sampling and link prediction techniques. We use mathematical and statistical tools to identify regions of dense connectivity — clusters — and optimally partition the graph. Each sub-community, tends to represent a group of users who share an interest. In the context of the 2016 US Presidential Elections, this aligns with political affiliation. Here’s what the model looks like with annotated sub-communities:
While there are two obvious groups that emerge from the model, Liberals and Conservatives, there’s a clear group of reporters as well as a few groups affiliated with Presidential Candidates. These are early days for the campaigns. Some are still in the process of announcing candidacy and all are actively hiring their teams. So as one can imagine, their networks and digital footprints are starting to take form.
Hillary is certainly one of the more visible candidates, yet her fans and supporters are still well embedded within the larger group of Liberals. Bernie Sanders and his supporters (Yellow nodes) are completely embedded within the cluster of Liberals. Similar yet somewhat different scale, Ted Cruz supporters are wholly intertwined within the Conservatives cluster of users (light red nodes) but comprise a large percentage from the whole. Carly Fiorina, who has a significantly smaller footprint online, is also fully embedded within the Conservative cluster. As these campaigns mature, their groups will be better distinguished at which point we’ll see them break away from the larger party rhetoric.
We take the graph above, a visual representation of the 2016 US Presidential Elections model, and feed it into the Scale Model interface, which gives us the ability to observe top users, hashtags, links, domains and images shared across the whole model, or within each sub-community.
By turning on and off specific sub-communities through the interface, we can compare, for example, messages bubbling up right now within the Hillary Clinton versus the Bernie Sanders supporters. #askingforafriend — Bill Clinton’s snarky tweet from yesterday — is still the dominant hashtag shared amongst Hillary’s supporters, while Bernie Sanders’ Periscope announcement on Public University tuition tops the list within his group.
Reaching Communities
We’ve provided access to six public communities — 2016 US Elections, Animal Lovers, Climate Change, Travel, TV Culture and US Politics — but have created many others over the past couple of months. If you’re interested in an additional topic, feel free to email us.
For some of our early targeting tests, we built a model that follows Stephen King fans and Horror film enthusiasts. Using this model we optimized our Twitter Ad targeting to The Derry Connection, a website that maps connections between Stephen King books. This produced very high levels of engagement on both the paid and organic measures — yes, some people will Retweet an ad!
A few weeks ago we helped BBC map out the political discourse around the UK General Elections and many months beforehand, partnered with Digg to cover the Scottish Referendum. Having built community models for both events, even though they’ve occurred in the past, the identified communities and sub-communities continue to actively share content, and continue to surface some important trends. For example, UK conservatives, right now, are obsessed with the mandated BBC Television Tax.
The Anti-Vaccine community is also an interesting one. Incredibly active and well-organized under a number of dominant hashtags, such as #SB277 and #CDCwhisteblower, the community shifts attention towards #VaxVote and #B1Less. The dashboard highlights top links, domains and media published by the group in aggregate.
The discrete items that bubble up are interesting enough. Yet identifying trends over time is key. It’s one thing to note a story popularly shared within a community of users, but another to identify the shift in frame — how do Ted Cruz supporters grasp topics around immigration reform and how do those change over time? How does the community evolve — some groups merge, others grow further apart. We’re still working on baking some of these insights into the tool.
It’s a beta.
Give the public models a try. We’d absolutely love your feedback. Do you find this tool useful in your day-to-day work? If so, in what way?
Shoot us an email at requests@scalemodel.com if you’re interested in a new model.
Or tweet at us — @scalemodelapp — with other suggestions!
![[in beta]](https://miro.medium.com/v2/resize:fill:64:64/1*v-NULuaXus0LWWmACqFrNA.jpeg){kind=small}