created with NetLogo
view/download model file: RandAndPrefNet.nlogo
This is a slightly modified version (by Lada Adamic) of the standard preferential attachment (Barabasi-Albert) model that comes in NetLogo's Models library.
In some networks, a few "hubs" have lots of connections, while everybody else only has a few. This model shows one way such networks can arise.
Such networks can be found in a surprisingly large range of real world situations, ranging from the connections between websites to the collaborations between actors.
This model generates these networks by a process of "preferential attachment", in which new network members prefer to make a connection to the more popular existing members.
The model starts with m nodes connected in a cycle.
At each step, a new node is added. A new node picks an existing node to connect to randomly. With probability gamma it will select a random node. With probability (1-gamma) it will attach with a bias. More specifically, a node's chance of being selected is directly proportional to the number of connections it already has, or its "degree." This is the mechanism which is called "preferential attachment."
Pressing the GO ONCE button adds one new node. To continuously add nodes, press GO.
The LAYOUT? switch controls whether or not the layout procedure is run. This procedure attempts to move the nodes around to make the structure of the network easier to see.
The PLOT? switch turns off the plots which speeds up the model.
The RESIZE-NODES button will make all of the nodes take on a size representative of their degree distribution. If you press it again the nodes will return to equal size.
If you want the model to run faster, you can turn off the LAYOUT? and PLOT? switches and/or freeze the view (using the on/off button in the control strip over the view). The LAYOUT? switch has the greatest effect on the speed of the model.
If you have LAYOUT? switched off, and then want the network to have a more appealing layout, press the REDO-LAYOUT button which will run the layout-step procedure until you press the button again. You can press REDO-LAYOUT at any time even if you had LAYOUT? switched on and it will try to make the network easier to see.
The gamma parameter determines the probability that a new node will attach at random, vs. preferentially according to the degree of the existing nodes.
The m parameters simply sets how many edges each new node has.
The networks that result from running this model in the limit of gamma=0 are often called "scale-free" or "power law" networks. These are networks in which the distribution of the number of connections of each node is not a normal distribution -- instead it follows what is a called a power law distribution. Power law distributions are different from normal distributions in that they do not have a peak at the average, and they are more likely to contain extreme values (see Barabasi 2002 for a further description of the frequency and significance of scale-free networks). Barabasi originally described this mechanism for creating networks, but there are other mechanisms of creating scale-free networks and so the networks created by the mechanism implemented in this model are referred to as Barabasi scale-free networks.
You can see the degree distribution of the network in this model by looking at the plots. The top plot is a histogram of the degree of each node. The bottom plot shows the same data, but both axes are on a logarithmic scale. When degree distribution follows a power law, it appears as a straight line on the log-log plot. One simple way to think about power laws is that if there is one node with a degree distribution of 1000, then there will be ten nodes with a degree distribution of 100, and 100 nodes with a degree distribution of 10.
Let the model run a little while. How many nodes are "hubs", that is, have many connections? How many have only a few? Does some low degree node ever become a hub? How often?
Turn off the LAYOUT? switch and freeze the view to speed up the model, then allow a large network to form. What is the shape of the histogram in the top plot? What do you see in log-log plot? Notice that the log-log plot is only a straight line for a limited range of values. Why is this? Does the degree to which the log-log plot resembles a straight line grow as you add more node to the network?
Try increasing the probability of random attachment (i.e. increasing gamma). What happens to the degree distribution. Is it still powerlaw?
Try increasing m, the number of connections each new node establishes. What do you observe about the network? The degree distribution?
Assign an additional attribute to each node. Make the probability of attachment depend on this new attribute as well as on degree. (A bias slider could control how much the attribute influences the decision.)
Can the layout algorithm be improved? Perhaps nodes from different hubs could repel each other more strongly than nodes from the same hub, in order to encourage the hubs to be physically separate in the layout.
There are many ways to graphically display networks. This model uses a common "spring" method where the movement of a node at each time step is the net result of "spring" forces that pulls connected nodes together and repulsion forces that push all the nodes away from each other. This code is in the layout-step procedure. You can force this code to execute any time by pressing the REDO LAYOUT button, and pressing it again when you are happy with the layout.
Both nodes and edges are turtles. Edge turtles have the "line" shape. The edge turtle's SIZE variable is used to make the edge be the right length.
Lists are used heavily in this model. Each node maintains a list of its neighboring nodes.
See other models in the Networks section of the Models Library, such as Giant Component.
See also Network Example, in the Code Examples section.
This model is based on:
Albert-Laszlo Barabasi. Linked: The New Science of Networks, Perseus Publishing, Cambridge, Massachusetts, pages 79-92.
For a more technical treatment, see:
Albert-Laszlo Barabasi & Reka Albert. Emergence of Scaling in Random Networks, Science, Vol 286, Issue 5439, 15 October 1999, pages 509-512.
Barabasi's webpage has additional information at: http://www.nd.edu/~alb/
The layout algorithm is based on the Fruchterman-Reingold layout algorithm. More information about this algorithm can be obtained at: http://citeseer.ist.psu.edu/fruchterman91graph.html.
For a model similar to the one described in the first extension, please consult:
W. Brian Arthur, "Urban Systems and Historical Path-Dependence", Chapt. 4 in Urban systems and Infrastructure, J. Ausubel and R. Herman (eds.), National Academy of Sciences, Washington, D.C., 1988.
To refer to this model in academic publications, please use: Wilensky, U. (2005). NetLogo Preferential Attachment model. http://ccl.northwestern.edu/netlogo/models/PreferentialAttachment. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL.