The tension between seeing the specific in the general and gleaning the general from the specific is perennial in all endeavours concerned with ‘reading’ reality and (re)presenting it again towards their peers. It is required to bring some order to the ‘flow of phenomena’, but then again, to do so properly means to present an order that supports the reality inherent the chaos, not to erase it or hide it. This thesis grew out of the ambition I voiced in the creation of a webcollection of archived religious and spiritual websites: to honour the diversity of religious expressions in the Netherlands. But diversity is difficult to work with, especially when we keep in mind the researcher who may be using this webarchive perhaps many years in the future, when the intuitions and assumptions present at the time of its
construction have long melted away. What does this ‘diversity’ really mean, and what are the limits of applying that moniker to a collection of websites. True, websites are characteristically fluid in their purpose, figuring variously as profiles representing views of self to the world, channels of expressions containing opinion, knowledge and faith and tools of organisation, communication, aspiration and/or commerce. But that does not free them from being enmeshed in a technological network which is itself interwoven through-and-through with a culture and economy spanning far beyond the borders of a single European state. And, far from being a ‘cyberspace’ subject only to its own logic, neither are the subjects populating the Internet. In other words, there is underlying unity and structure within the diversity of websites and their content. But equally true is that a structure (such as squaring religions into neat denominational categories) exists on top of a diversity, which does not mind flowing beyond the borders of conceptualisations. Can the general and the particular then only be viewed in separate instances, like the eye must choose to focus on either the near or far?
The ambition in this thesis was to put various models of reality ‘in dialogue’ to see if the particular can speak through the structural. The specific question attached to it was: ‘in what ways can the computational modelling of religious websites support or nuance their a priori categorisation?’
Answering this question entailed combining one kind of model of reality with another: the ‘organic’, public-discursive model that attaches names to forms of religion, and the computational that produces highly specific overviews of data associated with said named religions. The practice of hyperlinking has been laid bare in two series of hyperlink networks graphs, and parts of the texts used on these websites have been processed to see if a religion’s name can accurately classify whether a site writes directly about said religion.
We can conclude that, yes, this combination of models has resulted not in argument and
confrontation, with one triumphing and the other discredited, but in a real, mutually rewarding dialogue.
Firstly, the algorithms have shown that our classification of religion has not come about arbitrarily. More often than not, the boundaries between religious groups are maintained when visualizing the ways websites link to each other. In fact, in the case of interreligious links, the higher levels of abstraction provide clearer clusters than the lower levels, suggesting that such higher categories as ‘Christian’ or
‘Buddhist’ do truly envelop some regularity shared between the various subgroups that it represents.
The same emerges from the classifier experiment, the result of which is a high confidence that Christian sites share enough of a common language to be correctly separated from non-Christian sites.
In fact, the way the various clusters in the link graph are laid out spatially seems to correspond with the how these religions could be said to be conceptually apart from each other. As such, the various Christian subclusters are all adjacent to one another, while the spiritual and new religious movement-cluster is separated from Christianity by a Buddhist movement-cluster; a religion which is popularly seen as
78 straddling the line between traditional religion and modern spirituality. Something similar can be seen in the second series of graphs, which plot the corpus in their neighbourhood of outlinks. Many religions and spiritual movements are thought of as ‘alternative circuits’, operating beyond the discourses and
avenues of the social mainstream. And indeed, relative to the cluster of socially central sites, many of these peripheral movements hover at a distance, connecting only to a cluster of globally central, one could say ‘neutral’, web-infrastructural sites and platforms. Essentially, these visualizations are showing that the data associated with these religions confirm that there is some truth to the narratives
commonly associated with them.
And yet, if a critical eye questions and investigates the methods used to arrive at these surprisingly revealing graphs, questions the operations which have been carried out on the data and whether any aspects of them have been omitted, then the other side of the dialogue manifests.
“Buddhism’s” position as intermediary is an example. Its position between Christianity and Spirituality is given a stronger veracity by the graph than it may really have, simply because the direction of ties is ignored. What we may be looking at is Spirituality linking to Buddhist sites but being themselves ignored, while Buddhist sites for their part link to Christianity without the favour being returned. Significant characteristics about the nature of relationships are thus omitted for the sake of easier interpretation.
This can be a pitfall leading to misunderstanding if the researcher is not aware of what the data looks like and what an algorithm does with it, or deigns to explicate it to readers. But if this awareness is there, any result can be questioned in light of the knowledge of the operation. “Yes, the religious categories are meaningfully located, but I am looking at a simplified representation.” Subsequently, the modularity of computer analysis makes it relatively straightforward to redraw graphs with different parameters, and ask follow-up questions. “What if I retain the direction of the tie this time?”
Indeed, the greatest value of these models is unlocked when they are part of a process of discovery rather than an end point. Yes, they do show structures latent in the data, but for reasons inherent in both the data itself and the algorithms, these overviews raise as many questions as they answer. And this is what the researcher should want, to have their assumptions challenged by different looks at reality. The beauty of the graphical representations is that the unit – the node – is not dissolved into a sea of structure, but truly functions as a building block. And when there are blocks in a place where they are not expected, the viewer will notice them immediately. This utility and beauty is something that the classifier algorithm lacks, which transforms its data units into a textual-numerical form much harder to interpret by eye. Therefore, the ease with which today’s computers can
deconstruct and represent objects of study into a multiplicity of datatypes should trigger the itch to use multiple models, and net multiple angles.
The ancient Chinese Daoist text Dao De Jing writes something in the vein of: “going far means returning” (遠曰反).157 Modelling reality means moving away from it, first through datafication and then through processing and/or visualising of that data. The crux is then whether the outbound movement supports the return: can we re-engage the actual reality before the data with a deeper insight than before? We use these models not to get lost in successive levels of abstraction, but to return to the ‘real’
world of people and passions with a fuller knowledge of it. So while mathematically supported software that can turn hyperlink data into an explosion of relational colour, far beyond what our synapses could ever imagine, getting to the bottom of what, why and how remains a task for beings embodied in space, time and society. Whatever the sophistication of data and algorithms, without a ‘traditional’ researcher’s insight and knowledge of what that data means and represents, no model will come to fruition. And just
157 https://en.wikipedia.org/wiki/Fan_(Daoism)#Daodejing
79 like any tool only works in intimation with our innate biology and psychology, a model should not replace but support out our capacity for mental inferences, arguments and syntheses. They should facilitate completing and bringing into view the circle between the latent structures and the specific particulars.
And finally, the best models should be self-aware – that is, be transparent to us – that they are not any more ‘true’ that the data they process, despite their visual charisma and high status within the dataist paradigm. They owe their existence to the data, who in turn, must stand in service of truth.
80