• No results found

7 Results: Statistics, Graphs and classifier experiment

7.3 Link Network graphs, part 2: Neighbourhood of outlinks

61

62 from the clusters, meaning they are neither links unique to one site nor frequently shared by a larger group of nodes.

Figure 14. Full network, primary labels. Labels for corpus nodes. Node size for number of outlinks.

Figure 15, demonstrates just how prevalent this phenomenon – which we could term the ‘hapax-site’:

sites linked to only once – is. If we filter the non-corpus sites by selecting only those receiving at least two inlinks, the amount of nodes in the graphs is winnowed massively to just 6531, 11.70% of the initial 55.817. Little remains of the black clouds, although some sites, like reliwiki.nl, kruislinks.nl and

interestingly, scientology.nl, retain some of their retinue.

63

Figure 15. Full network, primary labels, filtered on > 1 inlinks.

Filtering the amount of non-corpus sites further down to those receiving at least 10 inlinks brings it down to a mere 876, or 1.57%. Subtracting 608, a mere 73 sites out of the initial 55.209 outer sites remain. Can they be said to be central to the interlinking activity of the corpus? For the following graphs, the

Fruchterman Reingold force algorithm has been used, which despite its age (having been published in 1991) draws the graph such as to enable some surprising insight. Figure 16 shows the result with the labels of the remaining outlinks drawn according to the amount of inlinks they receive. There are two clusters of grey nodes in the graph: a dense one in the center, containing the sites receiving the most inlinks, and a more spread-out one in the bottom-right corner, which contains sites at the lower end of link-reception. A minority of sites hovers in-between the clusters or around the central clusters. Also

64 notable are a number of ‘islands’ floating around the edge of the graph: sites which deign to connect to any of the most central sites.

Figure 16. Corpus links with primary colours. Out-links filtered on => 10 inlinks.

The site labels make it hard to mistake which sites are at the absolute center: the large social media platforms, Google, Wikipedia. This is little surprising. Figure 17 and 18 show further closeups of the two clusters. Disentangling the center brings a number of ‘variant’ URLs of the major platforms into view, as well as other major ‘infrastructural’ sites into view, such reddit.com, zoom.us, pinterest.com, dropbox.com, mailchi.mp, apps.apple.com and shopping sites bol.com and amazon.com. Interesting to note are also (semi-)governmental sites belangdienst.nl, rivm.nl and autoriteitpersoongegevens.nl. Such

65 generalizable sites are not the only ones in the center, however, as there are also a number of explicitly Christian ones, such as biblegateway.com, katholiekleven.nl, vatican.va, rkkerk.nl.

Figure 17. Close-up of the central cluster.

The closeup of the second cluster shows that it has its own substructure. In the center-left is what appears to be a center populated by major Dutch newspaper and websites of the government. The left side is rather strongly populated by Christian sites, both corpus and non-corpus. The bottom and right side, finally are populated by a wide array of ‘large’ websites, be it news publications, major universities, nongovernmental organizations, site hosters (angelfire.com), smaller platforms (such as academia,edu) or databases (imdb.com, web.archive.org), as well as some more Christian websites.

66

Figure 18. Close-up of the bottom-right cluster.

Having clarified what sites populate the two grey clusters, we can make a number of inferences about what the location of the corpus sites means about their connection to the outlinks. For instance, those sites which appear in-between the clusters will have to have connections to both the ‘platform’ cluster and the cluster of general sites, whereby those appearing more to the left will likely be more Christian, while those appearing to the right will be less Christian. A glance at the predominance of red (Christian) nodes in the directly beneath the central cluster and to the right of the general cluster seems to confirm this. Nodes that are enveloped or at the edge of the general cluster are likely to be less strongly

connected to the platform cluster. Finally, all those sites which populate the ring to the left, top and top-right of the graph’s center should be connected mostly to the central sites only, and in lesser degree to the general cluster. The more ‘northern’ a site is located, the more it is connected only to the platforms

67 and other major infrastructural sites. All this means that, if we discount the highly connected central Christian sites (due to whose influence a not insignificant number of the most linked-to sites are specifically Christian), we are left with three rough categories: sites connected to both platforms are generally popular sites, sites connected only to (some) of the generally popular sites, and sites connected only to the platforms, all of course with different intensities and specifics of connection. Having clarified this, the graph can be put to use in, for instance, comparing how different religions are connected to the two centers.

Figure 19. Protestant sites.

Figure 19 highlights all sites belonging to Protestantism: the most numerous category in the corpus. As mentioned above, a cluster of Christian sites is found directly below the central cluster and those outer

68 sites which are specifically Christian, and found around the bottom and bottom-left of the graph’s

general cluster. Moving to the left of this cluster is a selection of sites as numerous as this cluster: the sites are moving towards the periphery of the graph, which means they are lessening in their linkage to the general (Christian) sites and their position becomes more dependent on links to the central cluster.

Moving clockwise, this ends with a not insignificant number of sites which seem to be unconnected to the general sites, although there is a smaller ring of sites in the bottom half that contrastingly connects only to them. A tentative conclusion that could be drawn from such an analysis could be in the vein of:

Protestant sites tend to be generally, though far from uniformly, well connected to the most central outsites.

Figure 20. Roman Catholic sites.

69 Figure 20 shows Roman Catholic websites. The second largest Catholic denomination also possesses its own cluster in-between the two outlink cliques, located to the right of the Protestant cluster. There are also a number of sites directly above the central cluster, perhaps due to the presence of vatican.ca, perhaps because they are only just still connected to the general cluster. In any case, compared to Protestantism, Catholicism has relatively little sites in the peripheral, cluster-only-connection circle.

Figure 21. Orthodox and Oriental sites.

Figure 21 moves down one level of categorization displays the Eastern-Orthodox websites (Russian-Orthodox, Serbian-(Russian-Orthodox, etc.) and so-called Oriental-Orthodox Churches (prevalent in the Middle-East and the Horn of Africa). Indeed, this is where an unmistakable correlation between the corpus-only graph, and perhaps, analog reality, shows up. In the corpus-only graph, both groups were peripheral at

70 most, and this is clearly replicated in the current graph, where all but a few sites are found exclusively in the peripheral, topmost ring. This means their connection to popular Dutch and international sites is small indeed. They are connected to the broader internet only through social media and infrastructural websites: perhaps a reflection of these churches’ relative marginalization (be it self-imposed or not) within the broader religious landscape.

Figure 22. Gereformeerde and Hervormde Kerken sites.

Figure 22 shows categories from the tertiary level of categorization that, moreover, are known to not attach tremendous importance to centrality in secular, public life: the Dutch Reformed Churches (Gereformeerde and Hervormde Kerken). While a few sites do seem to fit this bill and appear as island floating around the major component, the majority of sites is located in between the two components,

71 indicating that they maintain both links to platforms & infrastructural sites and the generally used sites.

However, remembering that the left of the general cluster contained a number of Christian sites, and seeing how the Reformed churches do hover more towards this side, their connection to generally popular secular sites does not seem very strong. It would be interesting to filter out the Christian grey nodes and redraw the graph.

Figure 23. Jewish and Islamic sites.

Figure 23 moves back upwards to the primary level and shows Jewish and Islamic sites. Compared to Christianity, the distance from the general cluster is notable, even though here there are also a number of more central exceptions. Interestingly, more Islamic sites are found to the direct east of the main cluster, and more Jewish sites to its north-east, whilst its north-west – the point farthest away from the

72 general cluster – contains more Islamic sites. Despite a general lack of highly central connection, it is difficult without closer inspection to discern the reason for the relative position of the two religions.

Figure 24. Buddhist, Hindu and secondary traditional religion sites.

Figure 24 shows Buddhist, Hindu and ‘secondary traditional religions’ (such as Sikhism and Jainism). Here too, the central space between the clusters is rather sparse, and populated mostly by Buddhist sites.

Indeed, the while most Buddhist sites are not quite central, they do not appear fully peripheral either, with many appearing around the top edge of the central cluster. Hindu sites share this behaviour, aside from appearing in the bottom-left peripheral ring.

73

Figure 25. Spiritual and new movement sites.

The spiritual and ‘new movement’ categories contain an enormous variety of websites. Despite this, figure 25 seems to display in line with the conception of ‘spirituality’ as outside the mainstream circles of society, as the majority of nodes are found in the outer rings, relatively unconnected to the cluster of sites prominent in the mainstream. Due to the high level of abstraction that the categories of ‘Spirituality’

and ‘New movement’ represent, this is one of the situations where a more detailed categorization would be interesting to plot. Such a categorization would need a strategy different from the one I used,

however, which ‘ticked’ multiple categories based on the presence of a spiritual topic. To digest such an enumeration into a more specific generality that ‘spiritual’ would require applying another nomenclature than what my metadata contains.

74

Figure 26. Contemporary paganism sites.

Finally, the same can be said for sites belonging to strands of contemporary paganism, as visible in figure 26. Here too, only a few sites gather around the central space, with the rest hovering at a distance in the graph’s periphery.

This second section of graphs can be concluded with insights quite like those in the first.

Religious categorisation is again reinscribed with the value to predict the behaviour – that is, online interlinking – of the religious actor – a website and the people responsible for it – in the world. Given that accounting all the corpus’ outlinks nets one cluster of major infrastructural sites and another of general, mainstream ‘cultural content’ producers (interspersed with Christian sites), and subsequently, mapping the proximity of a religion’s nodes to either or both clusters results in images that accord with the highly simplified image of those religious as central or peripheral in Dutch society. The large Christian

75 denominations are most central, Orthodox churches, Islam, Hinduism and neopaganism are mostly peripheral, while Judaism, Buddhism lie in the middle. Spiritual/New movement sites have a small central subsection, while the rest is found throughout the outer rings. It has to be conceded here that the strength of definite statements about difference is somewhat lessened by the presence of Christian sites in both clusters, which biases Christian sites towards the middle and non-Christian sites towards the periphery. Nevertheless, the two ‘grey’ clusters are in no way dominated by them. Thus, we can use the language of the two mediatization theories to compare the way in which the internet is used to refer and link to the broader digital body politic. Indeed, it is somewhat safe to assume that those sites which only make use of major infrastructure like social media platforms are not central in the sense that they are actively engaged in public discourse. Of spiritual sites, orthodox churches and new religious

phenomenon, this behaviour is almost part of their allure and identity. Making statements about the inverse is not so straightforward, however. Inasmuch it can be said that the large platforms and infrastructural sites of the central cluster and the general sites of the other comprise the centre of the Dutch online landscape, or society in general, it would be inviting to conclude those sites in between represent a religious actor active in the middle of this landscape, were it not that the nature of the link remains unknown. It is just as possible for a central node to represent a website which finds important to refer to that mainstream without identification with it, perhaps as an instance of ‘critical linking’, as one of the top-10 ‘base outside links’ list, christianarchy.nl, hints at doing.

Therefore, the second common conclusion between this and the previous section is that these graphs provide an excellent staging ground for picking anomalies out of the visualized structure. Each graph displays a number of cases outside of the expected behaviour – sites as central Orthodox or neopagan sites. If inspection of sites hasn’t given raise to doubts about their normality within the religious context, perhaps because outwardly the site looks much like any other of its kind, than taking one data type and visualizing it in such a way can be enough reason to take another closer, critical look at parts of the corpus.