The Getty Provenance Index Remodel Project

Size: px

Start display at page:

Download "The Getty Provenance Index Remodel Project"

Imogen Rich
5 years ago
Views:

1 The Getty Provenance Index Remodel Project and Futures for the Study of the History of the Art Matthew Lincoln, Ph.D Data Research Specialist Getty Research Institute

2 Getty Conservation Institute Getty Foundation Getty Research Institute J. Paul Getty Museum

3 Getty Conservation Institute Getty Foundation Getty Research Institute J. Paul Getty Museum

4 Getty Vocabularies as LOD

The Getty Provenance Index (GPI) Databases Sample record: Lot 0352 from the GPI British Sales database in web interface @matthewdlincoln A set of 8+

5 The Getty Provenance Index (GPI) Databases Sample record: Lot 0352 from the GPI British Sales database in web A set of 8+ databases which together contain more than 1.5 million records Covers primarily Western European art (especially painting) from the 16th to the early 20th century

publication; they are built in the CuadraSTAR database system Databases comprise

6 The Getty Provenance Index (GPI) Databases The GPI was established in the 1980s as a book series GPI databases were designed to support the production of the print publication; they are built in the CuadraSTAR database system Databases comprise information from primary source documents such as archival inventories and auction

Challenges Current GPI data model is a partially-indexed quasitranscription Lack of data standardization Inconsistent/opaque relationships between records

7 Challenges Current GPI data model is a partially-indexed quasitranscription Lack of data standardization Inconsistent/opaque relationships between records Data is mostly about documents, but many users want data to be an index of events Image of Goupil & Cie Co. stock book, Livre no. 5, 1870

8 Conversion to Linked Data From records to semantic triples: Subject - Predicate Object Painting Artist Date Irises van Gogh, Vincent 1889 Guernica Picasso, Pablo 1937 Irises - created by - van Gogh Irises - created in

9 Things, not Strings Using http URIs enables disambiguation, which is essential for linking data sets across the web van Gogh Van Gogh Vincent van Gogh Vincent Willem van Gogh van Gogh, Vincent Also consider, different language appellations: Irises vs. Les Iris

10 @matthewdlincoln Events, not Records

11 @matthewdlincoln Production of Object

12 Transfer of Ownership of

13 History of Ownership of

16 But

17 Transcription or Reality? The PIR data is not always correct Sometimes contradictory (e.g. artist attribution) Is false information valuable?

18 Transcription or Reality? The PIR data is not always correct Sometimes contradictory (e.g. artist attribution) Is false information valuable? Historians: Yes - if it has a source!

19 From this model

20 ...to this. Modeling documents AND events means increasing complexity, reducing usability.

21 Additional challenges: Representing historical currencies decimalized, or specified down to individual coins? (pounds/shillings/pence issue) Editors entering qualifications like [?] years ago w/o documentation what kind of uncertainty did they mean? Old data entry fundamentally misunderstood some source documents -> systemic errors that can only be fixed by returning to originals

22 The negotiation continues Patrick

23 Data-Based to Data-Driven New Avenues for Research in the History of Collecting and Art Markets Sandra van Ginhoven, Matthew Lincoln, Christian Huemer (in preparation, forthcoming People Objects Prices Space Time

24 Machine Learning for Historical Interpretation Variables ML Algorithm Real process Abstract model Results

25 M. Knoedler & Co.

26 Train a machine classifier (random forest) to predict if Knoedler will turn a profit based on: Purchase price Time sitting in stock Type of buyer Year bought Genre of artwork Type of seller Nationality of artist Is the artist an old master? Is the artist still alive? Was the work purchased jointly with another dealer? Was the buyer a first-time buyer? Was the seller a first-time seller?

27 Random Forest (ensemble decision trees) Profitability is difficult, but not impossible, to predict. Our model correctly guesses either profit / loss about 75% of the time. Real process Abstract model Results

28 Before 1900 Early on, Knoedler had to sell quickly to improve chances at profit 100% 0% One year 100% Marginal change in chance of profitability 100% 0% 100% Five years but as they matured, the firm got more leeway. 100% 0% After 100% 2,000 4,000 6,000 Days in Knoedler's stock

29 buyer_type Artist Museum Dealer Collector Before After % Marginal change in chance of profitability 0% $10 $1,000 $100,000 $10 $1,000 $100,000 $10 $1,000 Knoedler may have sold frequently to other dealers but they didn t make much on those sales.

30 @matthewdlincoln Do painting value differentials between countries encourage international transactions? Use (Temporal) Exponential Random Graph Models (T/ERGM) to test net-based hypotheses What s the relationship between price differential and endogenous network effects like friend-of-a-friend? rich-gets-richer? small world ties?

PAINTING CHARACTERISTICS School of painting Size Support History painting Mythological theme Religious painting Allegorical theme Portrait painting Landscape theme only Still Life painting Landscape

31 PAINTING CHARACTERISTICS School of painting Size Support History painting Mythological theme Religious painting Allegorical theme Portrait painting Landscape theme only Still Life painting Landscape elements included Architectural constructions depicted Figures depicted Polished or glossy finishing Artist living at the time of sale DEALERS STRATEGIES Dealer Engravings done after the painting Previous collection/collector mentioned Associated work or artist or style The original painting is mentioned Paired painting Description in large font Textual classification framework based on attributes of paintings from their lot descriptions

Is the Fastest-Growing Segment of the Art Market the Cheapest?

AiA Art News-service Is the Fastest-Growing Segment of the Art Market the Cheapest? Prints and multiples, long an afterthought, are gaining momentum. By James Tarmy 29 januari 2018 10:00 CET In June 2017,