Visual network analysis helps sociologists, historians and economists examine social questions through digital technology.
The explosion of relational data in these researchers' workspaces forces them to find new analysis tools. Because their usual tools cannot grasp the connections within the data.
It is necessary to accompany them by providing a methodological framework and tools to explore and interpret these data.
Which Networks Are We Talking About?
The term "web" refers to the different objects that we tend to confuse. We can distinguish at least three of them.
- A social network: A set of individuals or organizations connected by social interactions. Connections are inexhaustibly rich: kinship, affinities, transactions, etc.
- Mathematical Graph: It is a 'node' structure connected by 'arcs' and whose topological properties are studied by 'graph theory'. On the contrary, the links are stripped of all their essence.
- Relational data that we use in practice, that is, digital files that list both items and the links between them. Elements and links are often defined by metadata.
These three objects are complementary. We naturally want to apply the power of mathematical computation to relational data to examine social questions.
This is possible, for example, provided we are aware of inconsistencies between objects that compel us to reduce it to a data in order to calculate social bond.
Relational data available to researchers are diverse. These are sometimes the result of empirical observations, such as "sociograms" that visualize relationships among students in an elementary school classroom.
In other cases, when the link is not a reduction of empirical reality, the data are formal from the outset.
It's like a collection of websites linked by hyperlinks. It applies to digital networks as well as citation networks that bibliometrics has been working on for a long time.
We visualize networks to explore. But we prefer to resort to scientific evidence, other ways. In 1970, exploratory data analysis was formalized.
It has been stated that classical statistical tools are designed to verify hypotheses, which statisticians use to find them.
This is a biased usage. That's because the hypotheses proposed by the data are statistically valid, even if, by definition, they are completely false. In other words, we can construct an example for any false hypothesis where it is statistically valid.
We can come to two conclusions. On the one hand, we need tools adapted to the generation of hypotheses.
On the other hand, it is preferable to try to build scientific evidence outside of the data we discover to formulate our hypotheses. It is in this position that visual network analysis becomes efficient.
It makes it possible to explore complex data to formulate research questions to be tested separately in other ways: measurements or qualitative study etc.
To explore a network, you must first visualize it. That is, an image or mapping must be created and then interpreted. In order to draw the nodes, positioning must be done in the plan.
For this, we use a different variant embedding algorithm that produces similar but different images.
Beyond their variations, these algorithms work on the same principle. Nodes push each other as links pull them in, and the network evolves like a giant mobile until it stabilizes.
In the resulting image, nodes are usually close together when they are directly or indirectly connected.
We cannot translate the distance between nodes more precisely, as the algorithm must make an approximation to flatten the mesh in two dimensions of the image.
Clusters that we define visually can be interpreted as communities under certain conditions. We can quantify the clustering phenomenon with measures such as density or modularity.
Gaps between aggregates can be interpreted as opposition or competition. The picture shows the structural features of the network.