View Source

{excerpt}Painting iconography by RKD{excerpt}

h1. Iconclass

h2. Iconclass RDF
Iconclass have been doing quite a bit of work on semantic publication: []

Content negotiation:
- Semantic URL: [], redirects to one of these below
- HTML view: []
- RDF []
- JSON: []

- 52D1: [], []
- 11H(BASIL THE GREAT): [], []
- 71P411(+1): [], []

h2. Data Set
[] : dump of the database in RDF N-Triple format.
Quite non-trivial:
- 19MB compressed, 340MB uncompressed
- 39578 concepts (terms); {nf} fgrep -c "skos/core#Concept" iconclass.nt {nf}
- 198120 prefLabels (5 languages per concept); {nf} fgrep -c "skos/core#prefLabel" iconclass.nt {nf}
- 3389605 explicit triples (85 per concept) {nf} wc -l iconclass.nt {nf}
- will probably grow 5x to 17M triples after transitive inference

h2. Notes
- since the Iconclass notations (codes) use special symbols (parens, spaces, plus), be very careful about encoding the URLs properly.
(!)(!)(!) Must check that OWLIM's RDF loader decodes the URL before writing it to the repo. Eg these two URLs are the same:
[] and []
- the code is mapped to skos:notation as expected
- they use skos:inScheme <> instead of the simpler <> I expected
- we need to be very careful to select only prefLabel in "en" since the full data includes multilingual labels (5 languages)
- includes nice "decoding" of the notation into descriptive dc:subjects.
Eg for [] "Susanna, Joakim's wife, walking in the garden; the two elders spying upon her"
the RDF [] includes:
<> dc:subject
"Daniel"@en, "Daniel 13"@en, "Old Testament"@en,
"Susanna"@en, "bible"@en, "elder"@en, "garden"@en,
"intervening"@en, "prophet"@en, "spying"@en, "walking"@en .

h2. Combinatorial Explosion
"This datafile does not contain all the expanded notations including keys.
There are about 1.3 million notations in ICONCLASS (with all keys and children full expanded).
The size would mushroom, so we are trying to figure out a better way to also make the expanded data available."

In other words the 39.5k "base" notations in the data file comprise only 3% of the total possible number.

What are these keys and where does this explosion come from?
- Apart from the basic notation, Iconclass can include a number of qualifiers in parentheses.
- eg 71P411 "Susanna, Joakim's wife, walking in the garden; the two elders spying upon her"
can be expanded to 71P411(\+1) "Susanna, Joakim's wife, walking in the garden; the two elders spying upon her (\+ Holy Trinity)".
- these qualifiers are hierarchical, eg at [] you see that (+1) is further subdivided to
(+11) (+ God the Father)
(+12) (+ Christ)
(+13) (+ Holy Ghost (as dove))
-- and if you click on (+13), you'll see it's further subdivided to
(+131) (+ Holy Ghost in other form)
- the main key determines the applicability of specific qualifiers
- the qualifiers are not globally unique.
Eg at [] you see that +11,+12,+13 mean a completely different breakdown (not related to the Holy Trinity)

Not only the data file skips a lot of the combinations, but the included combinations are missing some of the data of their base terms.
Eg [] does *not* have the dc:subjects listed for []

h2. Iconclass mappings
Hans Brandhorst [] on July 29, 2013: we are presently creating mappings between Iconclass and other vocabularies.
- A first mapping I undertook is between Iconclass and the (Dutch) keywords used by the Netherlands Institute for Art History.
- I've started a mapping with Library of Congress's Subject Headings on the basis of the use of LCSH by the Wellcome Institute. (More interesting perhaps for YCBA)

h1. Iconclass in ResearchSpace

h2. Iconclass for RKD Paintings
ResearchSpace includes a few Iconclass codes for RKD paintings. We made up local URLs since we didn't have Iconclass URLs.
<obj/2926> crm:P65_shows_visual_item <obj/2926/image>.
<obj/2926/image> a crm:E38_Image;
rso:P129_has_iconclass rst-iconclass:_71P412; # Susanna bathing, usually in or near a fountain and sometimes accompanied by two female servants

rso:P129_has_iconclass rdfs:subPropertyOf crm:P129_is_about.
- rso:P129_has_iconclass is a business-specific sub-property of P129_is_about (see [Rembrandt Changes])
- the iconclass term is represented in both SKOS and CRM

h2. Iconclass in YCBA LIDO

Emmanuelle: Part of what I have been doing this summer is assign unique identifiers to vocabulary terms. As you know ICONCLASS is one of the vocabularies that YCBA uses.
- Vladimir: There's no "iconclass" in the 13 LIDO sample files I have. Should we increase the sample set to include such examples, or you need to regenerate the LIDO?

They provide URIs for their subject headings, which I could provide in our data rather than the unique identifiers.
- So instead of this in LIDO:
<lido:conceptID lido:source="ICONCLASS" lido:type="subjectConcept">52D1</lido:conceptID>
<lido:term>Apollo and Diana kill Niobe's children</lido:term>
- We could emit this:
<lido:conceptID lido:source="ICONCLASS" lido:type="subjectConcept"></lido:conceptID>
<lido:term>Apollo and Diana kill Niobe's children</lido:term>
- Vladimir: Sounds better. Lec would have to use (or make up) URLs in the RDF in any case.
Since the Iconclass notations (codes) use special symbols (parens, spaces, plus), be very careful about encoding the URLs properly.

h2. Iconclass in YCBA RDF
- add the concept scheme to [Meta Thesaurus|Meta-Thesaurus and FR Names]
URL = []
hasDomain = crm:E55_Type
searchable = 1
taggable = 1
label = Iconclass
comment = Iconography (Iconclass subject)
- use in the object:
<object/1234> crm:P62_depicts <>
- Two alternatives about Iconclass terms data:
-# We load the full dataset as described above. Unfortunately we don't have an effort allocation for loading the Iconclass data set (which is non-trivial) in RS3.7.
(?) Do you use any expanded terms that are not in the dataset? See [#Combinatorial Explosion]
-# Lec emits term data in objects (then we'll be able to display them).
Follow the RDF examples as provided in [#Iconclass RDF]

h2. Iconclass for Image Retrieval
Iconclass is touted as "a multilingual classification system for cultural content". Two excellent examples of this aspect:
- []: shows example images in a coverflow
- Rijksmuseum's Open Access collection: see Iconclass act as a retrieval tool

Iconclass codes are hierarchical, and image retrieval by code should be transitive. An examination of the latter site shows this is the case: search by the higher level code includes all subcodes, eg:
| *n* | *Iconclass* | *Subject* |
| 142 | [71P4|] | additions to the story of Daniel |
| 88 | [71P42|] | the story of the image of Bel (Daniel 14:1-22) |
| 2 | [71P426|]| the priests of Bel and their relatives are arrested; they are put to death afterwards |

To achieve search by Iconclass we need to do this as part of "Update FRs for Yale":
- Complement FR2 with P62 since Yale uses this shortcut of P67 (see [FR Implementation#Thing is/has Type- FR2_has_type])
- Add skos:broaderTransitive in addition to P127 in [FR Transitivity]