I propose below, after a short introduction about the basis of SNA and some examples which shows the potential of this tool, a transcript of tutorial given during a workshop of the first Digital Humanities summer school in Switzerland (June 28. 2013), and kept up to date.
1. Short introduction to Social Network Analysis
A network consists of two components : a list of the actors composing the network, and a list of the relations (the interactions between actors). As part of a mathematical object, actors will then be called vertices (nodes, in Gephi), and relations will be denoted as tiles (edges, in Gephi).
By left, you can observe a very simple social graph, with both lists explicited. Two attributes are attached to the nodes : a label (his or her “name”) and a numeric attribute (akin to the sex of people here, for example). In the edge list, “Source” and “Target” entries refer to the nodes’ numeric identifiers (Id).
In our example, the “sex” attribute determines the color of the nodes. The size of a node depends on the value of its “degree centrality” (its number of connexions). The centrality measures are essential metrics to analyze the position of an actor in a network. They come in many variations, as shown at right (A = Degree centrality, number of connexions ; B = Closeness centrality, closeness to the entire network ; C = Betweenness centrality, bridges nodes ; D = Eigenvector centrality, connexion to well-connected nodes).
2. Graphs with GEPHI: some examples
3. Downloading GEPHI and the dataset
Download the application and both CSV files. This tutorial is based on the 0.8.2 Gephi beta version. If you encounter a problem due to a later update, do not hesitate to let me know.
The data consist of a random selection of Twitter users and their “followings” relations. The “Nodes” file contains the identifiers of each nodes, their label, a sex attribute and a random value that will be usefull to play with visualization tools hereafter. The “Edges” file contains a list of identifiers couples showing who follows who.
4. Importing the data into GEPHI
Run the application on your computer and create a “new project” in the start menu. In the Data Laboratory, click on “Import Spreadsheet” to open the import window and import your “nodes” file.
Specify that the separation between your data is expressed by a semicolon and do not forget to inform Gephi that the data you import is related to nodes, as demonstrated in this example (left). Then press “next” and fill the import settings form as proposed (right).
Follow the same procedure as for the nodes, but with the “edges” file downloaded above and by filling the forms in the following manner: specify the semicolon and inform Gephi that this time you import the edges. Fill in the last fields, and uncheck “create missing nodes”, because you’ve already imported them.
In the “Ranking” panel of the left column (top), select “Nodes” and the “red diamond” (size), then select “Degree” (rolling menu) and enter the minimal and maximal value (we propose 10-150). At that point, it’s possible to click on the “Spline” blue link to edit the shape of the spline (Be aware that linearly double the radius of the nodes is more than double the area because of the power function).
That’s the main part! While it is possible to play (and lose yourself) with various visualization capabilities, I propose a method based on this dataset. Start with Fruchterman Reingold (left column, bottom), and use the same values as in this model (10000 10; 10).
This visualization disposes nodes in a gravitational way (attraction-repulsion, in fact, as magnets). You’re already able to distinguish communities (more densely connected parts of the network). Let the function run until the graph is stabilized. Use the little blue magnifying glass (bottom left of the graph panel) to re-center the zoom.
Then, I propose to use the Force Atlas 2 (another layout algorithm) to disperse groups and give space around larger nodes. Be careful, the parameters you enter significantly alter the final appearance (proposition: Check “prevent overlap” and change “Scaling” to 10). Let the function run until the graph is mostly stabilized.
In the Ranking panel, choose the “color” sign to remove these sad shades of gray. As the nodes have attributes, you can color them regarding their “sex” attribute or their “value” (or simply again with their degree centrality).
Please note that if you used the “Spline” for the Nodes’ size, this setting will be used by default here (but can be modified now without interfering with your previous choice).
At the bottom right of the graph display, you find a little sign which allows you to develop a new panel. In “Label“, check “Nodes” to add their labels to your nodes and set their font/color/size… If wanted, you can click on the “Configure” link to set the data you want to get displayed.
For privacy reasons, the names are all displayed as “Names” in this dataset. No doubt you understand. Moreover, as the data set has been built (and not collected in a rational manner), the data can not be subject to interpretation.
Go to “preview” for trimming the final details. Unlike during previous stages, changing settings in this menu is reversible, and do not affect the structure of the graph. In the this screenshot, you will find a suggestion of settings for a good rendering. Be aware that due to its large size, the graph may take a few seconds to update after each change (click on “refresh” to apply the changes).
At the bottom of this preview column, you find an export link. Note that exporting in .png produces figure with a poor resolution. You may want to opt for .svg or .pdf, which have the advantage of being modifiable by your own image/drawing software (I recommend the open source program inkscape for manipulating .svg files).
6. Other features
The visualization is only one step, network analysis often needs other mathematical means to provide the researcher with a satisfactory result. Feel free to explore the “Statistics” menu (right), for example by playing with degree measures, density, path length, modularity.
A network contains internal subdivisions called communities. There are methods that permit to highlight these communities, which depend on the comparison of the densities of edges within a group, and from the group towards the rest of the network. More here!
In the right column of the “overview” page, click on Statistics/Modularity/Run to display the modularity window. Choose a resolution (between 0.1 and 2), click OK and close it.
The next step takes place in the Partition menu situated in the left column. Select “Nodes” and “Modularity Class” (rolling menu). You will be then able to modify the colors attributed to the detected communities by clicking on them.
Do not hesitate to repeat this operation with many “Resolutions” ! If you decide to do so, you must deselect and reselect “Modularity Class” in the left column, and refresh color calculation.
I hope this tutorial has been a way to whet your curiosity to go further in social network analysis, and I am delighted to see your accomplishments!