One of the things that interests me in social interest networks is the extent to which we can generate quick sketch maps of notable folk in an interest area, or generate quick profiles of the shared interests of a group of users. So here are a couple of doodles around the idea of social media triangulation, generated using Twitter, where I try to get an impression of the shared interest space of two or more users are located in .
The first attempt is a sample/sketch of the common friends three Twitter users (@clhw1, @melissaterras and @annewelsh) from the same institution (UCL),
The sketch is generated by grabbing the friends list for each using, constructing a directed graph of the folk they follow (from each person in the list to each of the people they follow), then filtering the graph to show nodes with outdegree > 0 or in-degree 3 (note to self: automate an option for this based on len(userList)):
If you know the identities of the folk identified in this sketch (I don’t) it may or may not be meaningful in terms of the names that are collected there, and maybe any names you might expect to be there that are absent… One obvious next step in profiling the shared interest area of the three originally identified users would be to generate a word cloud across the biographical descriptions of each of the folk identified in the sketch.
The second doodle is inspired by a James Allen post – What you will see Lewis Hamilton doing next:
“Lewis Hamilton has had a relationship with sports clothing brand Reebok since the early days of his F1 career and he is set to become one of the key faces in a new multi million pound campaign for the brand, set to launch in March.
According to Marketing Week, Reebok, which is part of the Adidas group, has identified that many people like to treat fitness and the act of getting and staying fit as a sport in itself and they are going to market it that way, using Hamilton and other sports personalities. It will be interesting to see how he is positioned”
Hmm… which got me wondering – if I grab the friends of a sample of the followers of @lewishamilton, and the friends of independent sample of the followers of @reebok, graph the result and then filter the net to show folk who follow both @reebok and @lewishamilton, who tends to be followed by these common followers? That is, given independent samples of followers of the followers of @reebok and @lewishamiliton, retain any folk in either sample who also follow the other target account and then map who this set of people follow to any significant extent?
Using a sample size of 197 random followers of @lewishamilton and an independent sample of 197 followers of @reebok, the number of individuals that followed both accounts was minimal, which is maybe not surprising given the small sample size and large follower counts.
However, if we relax the filtering constraint a little and instead plot the intersection of the (undirected) degree 2 egonets of @reebok and @lewishamilton, here’s what we get:
This graph was generated by plotting the friends network of 197 random followers of @lewishamilton and 197 random followers of @reebok and then applying the following filter:
That is, we filter the network (treating it as an undirected graph) to show all the people who are a within a network distance of two of both @lewishamilton and @reebok. Note that this does not mean that folk in the graph follow both @lewishamilton and @reebok, as this example shows [need a diagram!]: suppose A and B follow X, C follows Y, and D follows X and C.
Treating the edges as undirected, D is within a distance of 2 of both X and Y, but does not follow both X and Y directly.
A question that now comes to mind is this: does the “within distance 2” map have anything meaningful to say about the relative positioning of @lewishamilton and @reebok, or does the method of construction/filtering of the graph just produce meaningless noise?