From February 9^th to May 10^th 2015 I had the wonderful opportunity to be a fellow at the Harvard-Smithsonian Center for Astrophysics for the purpose of producing, with the SAO/NASA ADS (Astrophysics Data System) bibliographical data, a statistical study about the citing behavior of French astronomers. While the goal of my visit at the CfA was clear to me, many of my friends and relatives were rather surprised that someone coming from the very Humanities-centered École Nationale des Chartes and without any scientific knowledge other than high school mathematics would be invited to work in the astronomy and astrophysics disciplines. They were very right to be surprised as, even to myself, this space travel had absolutely nothing obvious. The basic purpose of my visit being, I concede, to learn data sciences and data visualization, I nevertheless explored further than those very practical skills and even discovered in myself a latent vocation of astrologer.

Epistemology

“Astronomy has always been a science of catalogs : catalogs of stars, galaxies, quasars, and planets.¹” This assertion of astronomer Joshua E.G. Peek is probably the first truth that I learned while working at the John G. Wolbach Library. If indeed at the beginning was the word, the word of my beginning was the one of the august and late Williamina Paton Stevens Fleming, whose journal² transcription was among my first tasks at the CfA. Williamina Fleming, hired in 1881 by Professor Edward Charles Pickering, discovered no less than aproximately 380 objects by classifying and cataloguing stars later published in the Henry Draper Catalogue. Amazed by this vivid record of astronomy in the making, I was very proud to consider myself a kind of continuator of her work by producing transcription templates of some manuscript notebooks : what better proof of the inherent human nature of Information Technology than the process of making computer readable data handwritten by a woman computer ?

Proud, yes, but also helpless : it is indeed very difficult to produce a template of a table when, to a non-astronomer’s eye, the metadata tend to confuse itself with the data it is actually pointing at (see figure below)! Thus another assertion was de facto verified, as the beautiful essay entitled Life and Death of Metadata³ puts it :

“(…) it is quite apparent that metadata are neither universal nor transparent, as various open data initiatives would have us believe. They are constructed for local purposes and audiences. (…) Indeed, as Bowker and Star documented in their classic work on classification, metadata are cultural, not natural. They are created, maintained and eventually discarded or replaced in craft practices that vary over time”.

Figure 1. Numbers, numbers everywhere!

Harvard Variables (noted as “HV.” on the upper left of the picture) are house made metadata to designate variable stars observed at the turn of the 20^th century. As explained to me by Harvard’s Professor Emeritus of Astronomy and the History of Science Owen J. Gingerich, each object was given a consecutive number as it was discovered on photographic plates at the Harvard Observatory. It is no coincidence that the expertise of no less than an eminent historian of astronomy would be needed to understand manuscript data. “The data, in and of itself, is only valuable [for] somebody who understands its significance” explains Peter Del Tredici in the essay previously cited. Data (or “capta”, as Johanna Drucker would call it) is a human artifact produced in a particular context and for a particular purpose; once traces of those elements are lost, the inherent value of data is lost, too. Indeed, astronomical data is to the astronomer as archives are to the historian: a record of a past object from which the scientist builds a sense meaningful to the very specific community being addressed. As the archivist’s task is to select among the huge amount of human activity records what documents will be most interesting to future researchers, a space telescope’s task is to focus on some part of the immensity of the cosmos according to the purposes of the most promising research proposal soliciting it. Moreover, data from telescopes and satellites need to be interpreted in order to be meaningful, whether through the mere transposition of a light spectrum in visible-to-human-eyes colors or through the more elaborate choice of a particular red to describe the behavior and temperature of a solar gas. This, in my understanding, testifies to the very symbolic nature of data as both being part and sign, substance and conventional construction of the object they capture. What, then, would better suit such a symbolic object than the very artificial projection of data visualization? “Indeed,” the author of Life and Death of Metadata writes, “visualizations are no more than visual metaphors, translating various kinds of quantitative data into spatial and geographical form.” Such a fact is very fortunate, as learning data visualization was one of the purpose of my visit at CfA.

Methodology

It would not come as a surprise to say that data science looks a lot like cooking. How do you make English pudding? How can we quantify French astronomer Daniel Egret’s suspicion that astronomers from a particular country tend to cite colleagues of the same nationality? And if we had to pick France as a country for this study, would papers written by mostly French collaborators tend to be cited by papers also authored mostly by French people?

Step 1: building a python function that would request bibliographical data of approximately 8600 records from the ADS API (Application Program Interface). Then, for each of those records, making another request to the ADS, mostly to get affiliations data of papers citing the records obtained from the first request (see below).

Figure 1. Data cooking, step 1

In order to make things simpler for you, reader, let us call the papers obtained from the first request “original records”, and the publications obtained from the second request “citing records”.

Step 2: computing the proportion of French authors in the affiliations of each of the original records, and then computing an average of the same proportion for the group of paper citing each of the original records. Thus, for each original record corresponds a proportion of French authors, and also the average proportion of “Frenchness” of the set of papers citing it. Here, you get that table:

Original paper identification number	Percentage of French affiliations in the original paper	Average percentage of French affiliations of the group of citing papers
xxx	YYY %	ZZZ %
xxx	YYY %	ZZZ %
xxx	YYY %	ZZZ %

Step 3: Trying to make something tasty with the ingredients you have prepared. For example, just take the first and second columns of the table and compute the cumulative distribution of original papers according to their level of “Frenchness”, which gives you the following histogram:

Then take each of the thresholds symbolized by these bins and, for each of them, compute the distribution of citing papers according to their level of “Frenchness”: you end with a series of 20 histograms which, put together, draw a picture of the evolution of the proportion of French affiliations of the citing papers according to the level of French affiliations of the original papers being cited (…dear reader, aspirin is in the kitchen). Now, it is only a matter of comparing the distribution of “Frenchness” of the papers citing 0.0% French original papers to the one of the papers citing 0.95% French original papers, to observe a positive and distinct gap between the first and the later:

And if you really want to quantify the augmentation of the proportion of French affiliation of citing papers according to the augmentation of that same proportion for original papers, just look at the evolution of the means and medians of each of the twenty distributions we obtained earlier:

Et voilà, bon appétit!

While reading my poor prose, dear reader, you may have wondered why I was making much ado about something that we already knew by intuition. But talking about bias in citation behaviors isn’t such a trivial issue, when we know the importance of bibliometrics in today’s networked way of accessing information.

“Citation and social computing indices,” writes Ronald Day, “begin with social and cultural categories and create documents and persons (users as well as authors) out of algorithms of information and communication “needs” (Thomas 2012), which reflect these social and categories in political circulation”⁴.

Thus, bibliographical data is of a very political nature, especially when used to algorithmically rank documents or authors and ultimately when it is involved in decision-making processes. How, then, can we not be skeptical about the functioning of such content providers as Scopus, which aims at offering “authors profiles which cover affiliations, number of publications and their bibliographic data, references and details on the number of citations each published document has received” and then ultimately provides navigation means based on our very sociologically biased citation data⁵?

That access to scientific information – and access to information in general, as the PageRank algorithm is the basic tool of Google – is a sociologically and politically loaded matter should not be considered malum in se. Rather, the question of designing a navigation tool able to reveal and question this subjective nature while providing an efficient way of finding unexpected and valuable documents in the context of an attention economy is the one we should consider as a priority.

I noted above that visualization had a metaphorical character, as well as the data it is representing has a symbolic nature. This is how we should read, in my opinion, Ronald Day’s statement about domain visualization in sciences, which says that “maps of science are representations of representations within a convenient metaphorical vehicle (“atlas”, “maps”).” Then, if we want visualization to be one way among others to call attention to the allegorical character of bibliographical data, it is important, that this visualization would carry an inherent playful and enjoyable aspect in order to maintain a distance with its content.

For example, dear reader, will you find seriousness in my study, if I present you the citation data that I used as flower-shaped (see below)? If I tell you that each point represents a publication, each petal of a flower cites its center, and that the more a point is blue, the more the affiliations of the paper are French, what will you think of the fact that most flowers whose center is blue tend to have blue petals too? Will you directly jump to the conclusion that I wanted to show you, which is that the more a paper is French, the more it’s cited by very French papers? Maybe, at least, will you get a sense of the symbolic, rather than indicative and realistic, nature of data that my flowers are supposed to evince?

Figure 2. Flower-shaped citation data

Choosing a metaphor to display data produced by a bibliometric study is one thing, but what about doing the same for a navigation tool, as previously mentioned?

Results

A good thing is that access to information has always been metaphorically mediated, as Scott Weingart writes in his article entitled “From trees to webs: uprooting knowledge through visualization”:

“Still, we have found the division of knowledge into subjects, disciplines or fields a useful practice since before Aristotle. These divisions are often organized into metaphors, which, in turn, influence our understanding of knowledge itself. Structured or diffuse; overlapping or separate; rooted or free, fractals or divisions; these metaphors inform how we think about thinking, and they lend themselves to visual representations which construct and reinforce our notions of the order of knowledge”⁶.

Of all the possible metaphors of knowledge, the cosmos metaphor is the one that, in my opinion, most efficiently describes the shape of our information. I’m not writing that just because of my particular love for heavenly bodies (really, stars are as pretty as flowers!), but rather because of the general tendency of “the universe of knowledge [to] strive little by little to imitate the model of the planetary universe”, as Umberto Eco puts it in his essay entitled “From the Tree to the Labyrinth”⁷.

The most exciting part of my visit at Harvard was probably when, by chance, I discovered that Jeff Steward, director of Digital Infrastructure and Emerging Technology at Harvard Art Museums, had already used the cosmos metaphor to design new “playful interactive interfaces for navigating the museums’ works of art”⁸:

Figure 3 Picture taken from Jeff Steward’s “CollectionSolarSystems”

You can access a demonstration of the Solar System visualization here, and the recipe behind it is on github.

Using planets and stars as a classifying metaphor is truly not a novel idea: this indeed is what astrology was (I insist here on the past tense) all about.

As sciences historian Gérard Simon writes: one can “analyze the astrological zodiac as one of the parts of what Lévi-Strauss calls in The Savage Mind a system of transformations. With references to the cardinal orientations which constitute it (equinoxes and solstices), combinations of elements which characterize signs (water, air, earth, fire), designation of the previous and properties associated to it, and finally, the precise binary rotation between males and females or diurnal and nocturnal, the zodiac participates in a complex system of classification, allowing the encryption of all the natural phenomena and the transformation of some of them in symbols. No being can escape the opposition of male and female, nothing the prevalence of earth, water, air or fire: everything is thus classified, and as planets obey the same classification, they can symbolically represent in the zodiac the class of beings or things which is supposed to carry a property similar to them”⁹.

Indeed, astrology in my understanding, is not the logos about the stars, nor the logos of the stars, it is a logos made of stars, where each sky element embodies and points toward a group of objects of knowledge, and can be manipulated in a similar way to words in a sentence. This would be a much more efficient and enjoyable way to critically represent library metadata and to use them as a navigation tool. And the fact that astrology is still largely denigrated today as a knowledge tool is only helping me more in my definitive intention of not being taken seriously. Astrology is actually part of Roland Barthes famous mythologies:

“The variations imposed, or rather proposed by the stars (for this astrology is a prudent theologian, it doesn’t exclude free will) are mild, they never seek to upset your life: the weight of destiny affects only your appetite for work, your assiduity or reluctance, the likelihood of promotions, the acrimony or complicity of your relations with your fellow workers, and above all your fatigue – the stars insistently and wisely prescribe more sleep, always more. (…) Yet even if its issues are pure mystification, even if it merely dodges problems of behavior, in the consciousness of its readers it remains an institution of reality: it is not a route of evasion but a realistic evidence of the salesgirl’s, of the employee’s life conditions. Then what purpose can it serve, this pure description, since it seems to offer no oneiric compensation? It serves to exorcise reality by naming it. By doing so, it joins the ranks of all the enterprises of semialienation (or of semiliberation), which make it their duty to objectify reality, though without going so far as to demystify it. We all know well at least one other of these nominalist attempts: Literature, which in its degraded forms can go no further than to name the experience of our lives; astrology and Literature have the same mission of ‘retarded’ institution of reality: astrology is the Literature of the petit bourgeois world”¹⁰.

Visualization, as mythology, needs to be broken in order to be truly informative.

Acknowledgements

My special thanks goes to the NASA/SAO ADS for allowing me to come spend those three fantastic months at the Harvard Center for Astrophysics. Then I warmly thank Christopher Erdmann, Alberto Accomazzi, Edwin Henneken, Owen Gingerich, Rahul Dave, Alexandra Holachek, Megan Potter, James Damon, Louise Rubin, Katie Frey, Maria McEachern, Amy Cohen for having made of my stay an unforgettable experience.