 Welcome to this video. I'm Martin Grangeant and with my colleague Mathieu Jacomi, we want to question the inter-subjectivity of network interpretation. What do we see when we analyze a network? This is sometimes very personal, depending on the disciplinary tradition that guides us, our technical skills and our understanding of the different theories that underline network analysis. This maneuvering process is subjective by nature, which may seem paradoxical considering that the digital methods are often supposed to objectify our research topics. Yet, as Karl Popper argued, descriptive statements necessarily draw the validity from inter-subjective agreement. How much agreement does network analysis offer? To discuss this, we extracted a network dataset that we each analyzed on our own. We didn't know the data in advance and had given ourselves a maximum of two hours. We recorded our screens and then made a quick commented montage of the main moments of our process. All this is then compiled on a web page that allows us to compare our interpretative journeys. A few words about the data. We worked on a network of Wikipedia pages that link to each other. We harvested the European cuisine category by extracting all its pages as well as the pages listed in its subcategories. To do this, we used notebooks developed by Mathieu and available on his mapping controversies website. We then extracted all the links between the pages of this list and kept all the pages that are linked at least twice. The network contains 5,800 nodes and 300,000 edges. Let's move on to the first part of this process, which is taking the data, opening a network analysis software and trying to see what it's all about. So I'm opening Fee, opening the GXF file and making a few calculations on the degree just to understand what's the dates about, just to understand the distribution of this metric to have a better idea of the complexity of the network I have in front of me. And then I use a few visual elements to make sure that I distinguish the bigger nodes and the small nodes. As a first step in my process here, I use the Fushtemann Heingold Spacialization Algorithm. I like the Spacialization Algorithm because it puts all the nodes in a specific circle and then all the nodes are the same distance from each other. It's a very soft first directed algorithm, I would say, and I always like to start with something like that just to have an idea of the network. Then I will move on to something more strong in terms of the Spacialization Algorithm to show the cluster a bit more. This first moment is also the moment of the first exploration just to have an idea of the structure of the graph globally, to have a look at the main clusters, to have a look at the main groups. And also the moment to start the first calculation, especially here, the betweenness centrality just to have an idea of the main linkers in this graph. So I changed the color to have an idea of their spatial position. And then of course I have a look at the data itself just to see the names of the nodes and the Wikipedia pages that are linking different groups together, hence the high betweenness centrality. Here I stop on a specific node, the biggest one just to see its ego network or all the nodes that are connected to this node. And I pick a few different nodes just to see if the node that is in the middle is connected to every other node or if it's not the case here, it's not completely the case, which is even more interesting. And now let's look at some more peripheral nodes just to have an idea of the different pages that are linked to this main red dot in the center here. Then I choose a more neutral color to go with the next part of our specialization process, which is applying for Sotlus2 here to give the clusters a little bit more space and to let the graph arrange around this new spatial referential. So here you see that the network is very highly clusterized. There are very specific groups in the periphery and in the middle a very large soup of nodes that are very hard to see because everything is entangled so that's why I'm using different visual parameters here to try to make the edges less visible and to have the nodes level appear better just to be able to explore it a bit more.