So, in addition to spatial models we just discussed, agent-based modeling is also found in a lot of applications recently to the study of social networks.

So, this is a paper I wrote with Forrest Stonedahl, he is at Augustana College and Uri Wilensky who you know is the architect of the NetLogo language and is at Northwestern University.

In this paper we were interested in understanding what is the best way to spread a social contagion through a network.

There has been a bunch of other papers on this, but we wanted to look specifically at the question as to whether or not local network properties were important.

And so what you see here is actually a realization of an actual Twitter network, that we had collected from Twitter data[?] via the API, of 1000 nodes.

And what we found in that network was that if I do a bunch of simulations using data-based modeling was

the contagion spread quickest through that network, when the people who were targeted with a message that you were trying to spread through the network had a lot of friends but their friends did not know each other.

So there were kind of these boundary spanning individuals who could spread the message through the entire network.

Forrest also built a nice 3D visualization of this that I will play now.

So what you see is the spread of this purple head idea, as we called it, throughout the entire twitter network we had collected which was 1000 nodes.

And as you can see, even in the small sample of Twitter, there are definite clusters of individuals.

And the idea spreads quickest if it is able to find the centers of all those clusters and then spread out through all of them.

It does not work if it is constrained to only one cluster.

So, in another paper I was involved with, where we were using agent-based modelling, we used ABM to try and infer a social network that we did not know the properties of.

So here is the basic idea: We took some facebook data about app adoption.

So we know how quickly on each day apps get installed. Right?

And then we tried to simulate what a bunch of different facebook networks would look like, given they would have those kind of app adoptions.

In other words, we were trying to figure out if we manipulate the topology of the underlying network, how does that effect the adoption pattern that we see.

And then we try to see, using a Bayesian method, which network was most likely to generate the adoption patterns that we saw on the facebook apps.

And interestingly enough we came up with a preferential attachment like model, which we will talk about later, is the most likely topological structure of the network, and with a fairly low density.

And you might think, why would it have a low density? Everyone knows everyone on facebook.

But in fact that is true, you know to a certain extend, but the people that you trust to recommend apps to you is probably a much smaller number.

And so that network probably has a much lower density than the overall facebook network.

So this is a piece of work that I have been working at very recently with Manuel Chica, who is a researcher in Spain.

And we have been working on trying to understand how do use agent-based modelling to help managers actually make decisions about word-of-mouth programs they are building.

And we call this a decision support system, which is a term that is often used in the literature when you are directly helping to try to make a decision.

So, what we do was, we use some of the previous research we had done on agent-based modelling, viral marketing and word-of-mouth programs.

And we got some data from some massive multiplayer online game, and we looked to see how adoption of premium content by one user seemed to spread through the network.

We then built a model that simulated that spread, an we allowed inputs to the model, such that managers could ask questions like

'if I wanted to provide incentives to talk about premium membership or to encourage other people to become premium members, who should I target in my overall network to do that?'

In this slide I show you some of the results of those targeting policy decisions, but it is also showing you the network of users that we were actually working with.

And here is that network again, but now with running some - in this case it is actually the real data, not the agent-based model -, but showing the possible connections of influence that happen within this network.

And we feel that is a powerful new use of agent-based modelling. We are actually using the tool to help people make real decisions about what policies to implement and how to understand those policies.

Now, I have spent some time over the last couple of talks, discussing applications of agent-based modelling in very complex settings to both spatial models and network based models.

Now, these are not the only places where agent-based modelling has some more complex and interesting insights to provide you with.

But they are two areas that I know about quite well, that I have spent some time in, so that is why I share those two areas with you.

I highly recommend that you look at a lot of the research that has been done, and I will use the Twitter feed also to post some of those papers out there.

And that will give you some more insight about complex applications of agent-based modelling.

You can also check out some of the websites out there where people post their models like openabm.org and the NetLogo modellings comment.

So that will provide you with additional insight to how complex these models can get.

So, thanks!