Research team: Kickstarting network analysis



Hello everyone!

Here is a quick post to kick start the network analysis.

The EdgeRyders data is rich in heterogeneous information and I would like to sketch together the kind of analysis we can build thanks to #futurespotters activities.

@Alberto, I am recalling the figures from our last discussion: the analysis will focus on about 100 users from many places, working together on an approximate 100 posts, with around 500 comments on these posts. However we have evidence that the activity on #spot-the-future ripples also on different campaigns, which we want to know also.

The first network analysis we are going to tackle is the collaboration of users within #spot-the-future, looking at how #futurespotters reply to each other on different posts.

We will create for this purpose a multiplex network of users in which an edge is a direct reply to another user.

To look at the “ripple” effect of #spot-the-future in other missions, with the same group of users, we will create a bipartite network (or affiliation network) of these people to all the posts they are participating in #EdgeRyders. If I noted it well, @Alberto, there is about 9000 comments to scan, right?

Later on, with the help of @Inga Popovaite , we’ll do further analysis… but I leave this for another post. Well, this is of course only the beginning of the discussion, feel free to join and enrich it!

Edgeryders LBG: one year checkpoint


Why do you need an affiliation network? You can simply:

  1. look into the STF groups and select all users that have been active on them
  2. look into the whole ER database and select all comments that have been written by and to these users. 
  3. this will be STF-induced activity
  4. by keeping track of the group ID you can tell which of the comments have stayed in the STF groups and which have spilled over to the broader ER conversation.
  5. the STF-induced activity will then be a part of the broader conversation, via the network itself. Studying its properties and relationship with the broader network is interesting in itself. For example: does Louvain find a reasonable approximation of the STF crowd?

This is an interaction network: Alice had an edge to Bob if and only if Alice has actually left a comment to Bob’s content. The nice thins about it is that it is straightforward to interpret. How would you use an affiliation network instead?


Yeah, affiliations are rich!

Of course we’ll do the analysis on this projection of the affiliation.

Imagine now we work with the multiplex network induced by the affiliation network. In this analysis, there is no doubt that the STF group will be the most tangled, but we can see among the other groups more tangled groups, which may also emphasize subgroups of the STF group.

The nice thing about the affiliation network is that it will allow us to look at both sides, groups and users.

For our meeting tomorrow, can we extract the first step, then I’ll start this right away!


It’s a plan

Ok, to tomorrow then!