Data Science and Digital Art History

Data Science and Digital Art History Лев Манович

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Data Science and Digital Art History 2015 Open MIND 10.11588/dah.2015.1.21631 2015-data-science-and-digital-art-histo 2015-data-science-and-digital-art-histo.txt, the project's staged copy 666ab98749c2aac985ef182363bb74e6608d06c4 open-access digital-humanities

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them method Quotation withheld because this record is not cleared for redistribution. computational analysis of large visual cultural datasets Quotation withheld because this record is not cleared for redistribution. review review Quotation withheld because this record is not cleared for redistribution. mechanisms Because a computer sees only low-level data (pixels, letters) and not meaning, analysts extract many additional features in the hope that together they carry enough information to bridge that gap. Quotation withheld because this record is not cleared for redistribution. mechanisms Human side-by-side comparison breaks down as collections grow, which is what forces the substitution of computed distance in feature space for perceptual comparison. Quotation withheld because this record is not cleared for redistribution. mechanisms The need to compute by hand constrained the founders of statistics to one- and two-variable methods, and that constraint still shapes how data is imagined today. Quotation withheld because this record is not cleared for redistribution. The chosen examples are deliberately biased toward the humanities rather than being a neutral survey of data science. what this limits: the choice of examples, declared as a specification rather than confessed: SINCE MY ARTICLE IS WRITTEN FOR HUMANITIES AUDIENCE, IT IS ON PURPOSE BIASED - MY EXAMPLES COME FROM HUMANITIES AS OPPOSED TO ECONOMICS OR SOCIOLOGY Quotation withheld because this record is not cleared for redistribution. A two-feature comparison of images ignores every other kind of difference between the works. what this limits: the two-feature comparison: THIS EXAMPLE DISREGARDS ALL OTHER TYPES OF DIFFERENCE: SUBJECT MATTER, COMPOSITION, COLOR PALETTE, BRUSHWORK Quotation withheld because this record is not cleared for redistribution. Even with many features there is no guarantee that the extracted features capture difference as humans perceive it. what this limits: any feature-based measure of similarity: EVEN IF WE USE MANY DIFFERENT FEATURES, WE CANNOT BE SURE THAT WE HAVE CAPTURED THE RIGHT INFORMATION TO QUANTIFY DIFFERENCE AS WE HUMANS SEE IT Quotation withheld because this record is not cleared for redistribution. Dimension-reduced plots are hard to interpret because the axes no longer correspond to single features. what this limits: any reading of the reduced plots: IT CAN BE QUITE CHALLENGING TO INTERPRET THE MEANING OF EACH DIMENSION once dimension reduction has been applied Quotation withheld because this record is not cleared for redistribution. Computational analysis has a precondition that the article says cannot be changed: the phenomenon must first be made into a finite set of objects and a finite set of features. what this limits: any computational result about a cultural phenomenon: the article states the phenomenon must first be represented as a finite set of individual objects and a finite set of their features Quotation withheld because this record is not cleared for redistribution. The comparison is confined to the chosen features, and everything else is set aside — which the article names as a limitation and then reframes as an advantage. what this limits: a similarity computed over selected image features: the article states it disregards subject matter, composition, colour palette and brushwork, and treats that as both a limitation and an advantage Quotation withheld because this record is not cleared for redistribution. The demographic and facial variables in the featured project were produced by paid crowdworkers and a commercial face-analysis service. what this limits: the age, gender and emotion variables in this project: the age and gender estimates came from Amazon Mechanical Turk and the facial features from a commercial recognition service Quotation withheld because this record is not cleared for redistribution. THE PAPER STATES THAT THE CENTRAL RISK CANNOT BE RETIRED BY ADDING MORE MEASUREMENT: 'even if we use many different features, we cannot be sure that we have captured the right information to quantify difference as we humans see it'. It then declines the obvious repair - 'producing a single correct map should not be our main goal' - so the limitation is accepted as a condition of the method rather than treated as a defect to fix. Quotation withheld because this record is not cleared for redistribution. the core analytic concepts of data science (objects, features, feature space, distance, dimension reduction) are imported wholesale as a framework for art-historical and humanistic study Quotation withheld because this record is not cleared for redistribution. the term "features" is adopted into digital art history in place of the sciences' "variable", and stretched to cover existing metadata as well as computed attributes Quotation withheld because this record is not cleared for redistribution. data data Quotation withheld because this record is not cleared for redistribution. phenomenon phenomenon Quotation withheld because this record is not cleared for redistribution. features features Quotation withheld because this record is not cleared for redistribution. objects objects Quotation withheld because this record is not cleared for redistribution. feature space feature space Quotation withheld because this record is not cleared for redistribution. phenomenon - AS INDEPENDENT: something that objectively exists regardless of how we study it, irrespective of which objects and properties we choose Quotation withheld because this record is not cleared for redistribution. phenomenon - AS CONSTITUTED: equal to the set of objects and properties used to study it, so that it is made by its representations and the conversations about it Quotation withheld because this record is not cleared for redistribution. feature selection - stated as BOTH a limitation and an advantage, because isolating features is what makes comparison on chosen dimensions possible at all Quotation withheld because this record is not cleared for redistribution. variable - used in opposite ways by the two fields the paper is joining: one isolated variable for understanding in statistics, many features for automation in data science Quotation withheld because this record is not cleared for redistribution. extends → Franco Moretti, 'Conjectures on World Literature' (2000) Quotation withheld because this record is not cleared for redistribution. features —operationalised-as→ feature space features feature space strength of the relation: 0.9 · note: The second constituting move, stated as a conceptualisation rather than a discovery: 'we will CONCEPTUALIZE our data table as a geometric space…' Similarity becomes distance because the table is read as a space. Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span says each feature BECOMES A DIMENSION of the feature space; operationalised-as is loose - part-of is what is stated. Quotation withheld because this record is not cleared for redistribution. principal component analysis principal component analysis Quotation withheld because this record is not cleared for redistribution. multidimensional scaling multidimensional scaling Quotation withheld because this record is not cleared for redistribution. mean mean Quotation withheld because this record is not cleared for redistribution. Classical statistics and data science use variables in opposite ways: statistics isolates one independent and one dependent variable in order to explain a phenomenon, while data science piles up many features in order to automate a cognitive task. Quotation withheld because this record is not cleared for redistribution. The feature-space concept lets the humanities' oldest method, comparison, be scaled up to very large cultural datasets. Quotation withheld because this record is not cleared for redistribution. Representing images in a feature space equates visual difference with distance in that space. Quotation withheld because this record is not cleared for redistribution. Automatic classification, probably the most common data-science application, is in the author's view the least interesting one for the humanities. Quotation withheld because this record is not cleared for redistribution. Dimension reduction makes the shape of a high-dimensional space visible but at the cost of interpretability, because each plotted axis stands for a combination of features rather than one feature. Quotation withheld because this record is not cleared for redistribution. On average brightness and saturation, van Gogh's Paris paintings vary more than his Arles paintings. Quotation withheld because this record is not cleared for redistribution. The images popularly associated with Impressionism make up only a smaller part of the artworks these artists created. Quotation withheld because this record is not cleared for redistribution. The Instagram tags show little interaction between two types: one group of users used generic tags, another tagged local, specific events. Quotation withheld because this record is not cleared for redistribution. data science data science Quotation withheld because this record is not cleared for redistribution. features features Quotation withheld because this record is not cleared for redistribution. data data Quotation withheld because this record is not cleared for redistribution. feature space feature space Quotation withheld because this record is not cleared for redistribution. dimension reduction dimension reduction Quotation withheld because this record is not cleared for redistribution. semantic gap semantic gap Quotation withheld because this record is not cleared for redistribution. coding coding Quotation withheld because this record is not cleared for redistribution. supervised learning supervised learning Quotation withheld because this record is not cleared for redistribution. average color saturation average color saturation Quotation withheld because this record is not cleared for redistribution. average hue average hue Quotation withheld because this record is not cleared for redistribution. 10.5 million — number of publicly visible Instagram images collected across New York City over five months in 2014 for the On Broadway project Quotation withheld because this record is not cleared for redistribution. 6,165 / 13,208 / 21,465 (5845 unique) — Instagram users, images and tags (and unique tags) collected in central Kyiv over 17–22 February 2014 Quotation withheld because this record is not cleared for redistribution. 200 — number of visual features the lab extracted from each Impressionist painting image before reducing dimensions Quotation withheld because this record is not cleared for redistribution. approximately 6000 — size of the French Impressionist painting dataset visualised with PCA Quotation withheld because this record is not cleared for redistribution. 3200 — number of Instagram self-portraits assembled and made filterable in the Selfiecity interface Quotation withheld because this record is not cleared for redistribution. 50,000 — size of the random Instagram image sample taken from each of Bangkok and Tokyo for the Phototrails comparison Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.11588/dah.2015.1.21631.