Linked Open Realities: : The Joys and Pains of Using LOD for Research

I presented this talk at the June 1, 2016 meeting of the CESTA (https://cesta.stanford.edu/) Graduate Fellows at Stanford University. What follows here is the translation of some very loose and informal notes into something more resembling a blog post.

Full text viewable at http://matthewlincoln.net/2016/06/06/linked-open-realities-the-joys-and-pains-of-using-lod-for-research.html

Transcript

1. Linked Open
   Realities
   Matthew Lincoln, PhD
   University of Maryland
   @matthewdlincoln
   June 1, 2016
   CESTA, Stanford University
   The Joys and Pains of Using
   Linked Open Data for Research
   

   View Slide

2. Image: @LegoAcademics CC-BY-NC-SA
   

   View Slide

3. Use Case: Print Production Networks
   ! LOD ! " LOD "
   1
   SPARQL encourages complex
   data expression and queries
   Composing SPARQL for an
   unfamiliar repo is challenging
   2 LOD is addressable
   LOD is just data on
   someone else’s computer
   3
   Different LOD repositories are
   linkable
   Content producers rarely use
   existing authorities/thesauri
   At the end of the day, I just need data
   @matthewdlincoln
   

   View Slide

4. Use case:
   Print Production Networks
   @matthewdlincoln
   

   View Slide

5. “Sculptura in Æs”, from Jan van der Straet’s
   Nova Reperta. Published by Philips Galle, c.
   1588-1605. The Metropolitan Museum of Art.
   @matthewdlincoln
   

   View Slide

6. Jan van der Straet
   Designer
   Jan Collaert I
   Engraver
   Philips Galle
   Publisher between 1588-1605!
   @matthewdlincoln
   

   View Slide

7. ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   Jacob Matham
   Claes Jansz. Visscher
   Hendrick Goltzius
   Karel van Mander I
   Jacques de Gheyn II
   Bartholomeus Spranger
   Jan Harmensz. Muller
   Harmen Jansz. Muller
   Jan van Londerseel
   Dirck Barendsz.
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   ●
   Claes Jansz. Visscher
   Gerard van Honthorst
   Theodor Matham
   Frederick de Wit
   Cornelis Danckerts I
   Jonas Suyderhoef
   Cornelis Visscher
   Michiel Mosyn
   Clement de Jonghe
   Cornelis van Dalen I
   @matthewdlincoln
   Example networks
   1580-1590 1640-1650
   Accepted for Journal of Digital Art History (summer 2016)
   

   View Slide

8. Between 1550-1750:
   Mining the museum for data
   @matthewdlincoln
   British
   Museum
   Rijksmuseum
   49,306! 46,730!prints
   3,592! 5,644!nodes: distinct designers, printmakers, and publishers
   76,697 ! 94,282 !edges: connections inferred from co-participation in an object
   Accepted for Journal of Digital Art History (summer 2016)
   

   View Slide

9. Use Case: Print Production Networks
   ! LOD ! " LOD "
   1
   SPARQL encourages complex
   data expression and queries
   Composing SPARQL for an
   unfamiliar repo is challenging
   2 LOD is addressable
   LOD is just data on
   someone else’s computer
   3
   Different LOD repositories are
   linkable
   Content producers rarely use
   existing authorities/thesauri
   At the end of the day, I just need data
   @matthewdlincoln
   

   View Slide

10. JSON vs. RDF/LOD
    @matthewdlincoln
    

    View Slide

11. @matthewdlincoln
    SPARQL
    Protocol
    And
    RDF
    Query
    Language
    

    View Slide

12. @matthewdlincoln
    http://programminghistorian.org/lessons/graph-databases-and-SPARQL
    

    View Slide

13. @matthewdlincoln
    http://programminghistorian.org/lessons/graph-databases-and-SPARQL
    SELECT ?artist ?painting
    WHERE {
    ?artist .
    ?painting ?artist .
    }
    

    View Slide

14. Writing SPARQL
    @matthewdlincoln
    

    View Slide

15. @matthewdlincoln
    

    View Slide

16. #!
    $!
    %!
    @matthewdlincoln
    

    View Slide

17. Unknown unknowns in
    graph databases: the
    case of the nationless
    Anthony van Dyck
    (1599-1641)
    @matthewdlincoln
    

    View Slide

18. Use Case: Print Production Networks
    ! LOD ! " LOD "
    1
    SPARQL encourages complex
    data expression and queries
    Composing SPARQL for an
    unfamiliar repo is challenging
    2 LOD is addressable
    LOD is just data on
    someone else’s computer
    3
    Different LOD repositories are
    linkable
    Content producers rarely use
    existing authorities/thesauri
    At the end of the day, I just need data
    @matthewdlincoln
    

    View Slide

19. SPARQL endpoint
    JSON API
    difficult to bulk download
    too unreliable
    scrape/download and
    create local copies
    Processing
    1.  cleaning
    2.  reshaping
    3.  network analyses
    4.  simulation
    Derivative datasets
    1.  Cleaned and filtered data
    2.  Analysis results
    Visualization &
    interpretation
    Disseminate
    @matthewdlincoln
    

    View Slide

20. @matthewdlincoln
    

    View Slide

21. @matthewdlincoln
    Image: Randall Munroe (www.xkcd.com/908) CC-BY-NC
    

    View Slide

22. @matthewdlincoln
    

    View Slide

23. Use Case: Print Production Networks
    ! LOD ! " LOD "
    1
    SPARQL encourages complex
    data expression and queries
    Composing SPARQL for an
    unfamiliar repo is challenging
    2 LOD is addressable
    LOD is just data on
    someone else’s computer
    3
    Different LOD repositories are
    linkable
    Content producers rarely use
    existing authorities/thesauri
    At the end of the day, I just need data
    @matthewdlincoln
    

    View Slide

24. What LOD/Semantic Web
    promised the cultural
    heritage world
    The LOD that the cultural
    heritage world has produced
    (at least so far)
    Image: Manu Cornet (www.bonkersworld.net)
    Image: Richard Cyganiak (www.lod-cloud.net)
    @matthewdlincoln
    

    View Slide

25. View Slide

26. Use Case: Print Production Networks
    ! LOD ! " LOD "
    1
    SPARQL encourages complex
    data expression and queries
    Composing SPARQL for an
    unfamiliar repo is challenging
    2 LOD is addressable
    LOD is just data on
    someone else’s computer
    3
    Different LOD repositories are
    linkable
    Content producers rarely use
    existing authorities/thesauri
    At the end of the day, I just need data
    @matthewdlincoln
    

    View Slide

27. Image: @LegoAcademics CC-BY-NC-SA
    @matthewdlincoln
    

    View Slide

28. Image: @LegoAcademics CC-BY-NC-SA
    @matthewdlincoln
    LOD desiderata from a researcher perspective:
    •  Visual interface for grokking an unfamiliar LOD repository
    (or remembering how a familiar one works)
    •  Guided SPARQL query construction (caution: might not be
    able to cater to full range of SPARQL complexity)
    •  Repo audit: for a given rdf:type of subject, what
    predicates are usually present? When are these predicates
    missing for highly-connected subjects?
    •  Google OpenRefine, but for fuzzy matching against
    different authorities
    

    View Slide

29. Image: @LegoAcademics CC-BY-NC-SA
    Matthew Lincoln
    matthewlincoln.net
    @matthewdlincoln
    

    View Slide