Intro

In order to explore the viability of different matching approaches, we decided to do some matching experiments and see what we can see.

• We'll use the web search interfaces of the target data sets described below and do manual searching.
  Please add web links below.
• For best results with automated matching, the target data sets wil/l be loaded locally.

Terms to match

Please add a random sampling of data below

BM People-Organizations

Target: VIAF

• Incorporates Getty ULAN, but has fewer fields (eg doesn't know person-person relations)
• VIAF has a loadable data set and Vlado has explored it somewhat
• Note: don't expect to find in VIAF many people that are not authors/artists. Eg
  • Catherine Talbot is in VIAF. She's a writer
  • Hon Arabella Talbot is not in VIAF. She's the wife of some dignitary. Presumably a collector or similar.
    Has no relation to any of the 115k objects, but may have some realtion to the 2M objects.
    http://collection.britishmuseum.org/id/person-institution/165252

Rembrandt in ULAN

Let's see what ULAN has on Rembrandt:
http://www.getty.edu/vow/ULANFullDisplay?find=rembrandt&role=&nation=&prev_page=1&subjectid=500011051

• Important ground-facts that we should use:
  • full list of alternative names (with useful scope notes)
  • nationalities
  • roles/professions
  • gender
  • birth/death dates and places
  • period and place(s) of activity
• Related people (family relations but also patron/ teacher/ follower...).
  • This is a great resource, with definite info, and we should strive to use it.
  • It is mathematically closed (maps from people to people), so once we have a match we should leverage it to generate other matches.
  • It does not make us solve a harder task (matching Works), as a map person->Works would do.
• NO list of works. Getty's only now working on CONA, but I guess it will be many years
• Lists of refs to biographies, references, subject classifications, contributing institutions.
  This established provenance, but not data useful for matching

Rembrandt in VIAF

Let's see what VIAF has on Person = Rembrandt Rijn
There are 3 uncorrelated clusters for Rembrandt, plus 6 more clusters about other people.

Let's see the main cluster:
http://viaf.org/viaf/64013650/#Rembrandt_Harmenszoon_van_Rijn,_1606-1669

It does list 86 works under "Uniform Title Links".

• A lot of these are not coreferenced e.g. "Die Anatomie des Doktor Tulp" vs "Anatomy Lesson of Dr. Tulp"
• See how many names about a single work:
  http://viaf.org/viaf/175917895/
  Bethsabée au bain tenant la lettre de David‏
  athsheba reading King David's letter‏
  athsheba with King David's letter‏
  ethsabée tenant la lettre du roi David‏
  Bathseba‏
  Bathsheba bathing‏
  Bathsheba with David's letter‏
  Bathsheba reading King David's letter‏
  Bathsheba with King David's letter‏
  Bethsabée au bain tenant la lettre de David‏
  Bethsabée tenant la lettre du roi David‏

BM People query

BM People sample

Alfred Smith

• BM data:

  id	prefName	altNames	activity	gender	nationality	birth	death
  78705	Alfred Smith	Smith, Alfred	clockmaker/watchmaker	Male	British	-	-
  195294	Alfred Smith	Smith, Alfred	architect	Male	British	-	-

• ULAN data:

  id	Uniform name	Roles	Gender	Nationalities	birth	death
  500007285	Smith, Alfred (French artist, 1853-1942)	artist,painter	male	French	1853	1942
  500007285	Smith, Carlton Alfred (British painter, 1853-1946)	artist,painter	male	British	1853	1946

In this case we can do reliable matching (NO match) based on the structured data: Activity does not match Roles.

BM People to ULAN

BM Places

Target: there are 3 competing data sets and no clear winner

• GeoNames is the best source of modern names.
  But museum data sometimes goes below the place level (eg a certain Cathedral or Theatre)
• Pleiades for ancient names
• Getty TGN

BM Places query

Note: you'll get some Archaic names if you set offset 100.

BM Places sample

BM Objects

Target: Getty AAT, Objects facet
Auxiliary source: Wordnet.

BM Objects query

BM Objects sample

BM Materials

Target: Getty AAT, Materials facet
Auxiliary source: Wordnet.

• Eg see sysets and semantic relations for cotton

BM Materials query

BM Materials sample

BM Materials to AAT

BM	BM prefLabel = altLabels	BM broaders	AAT	AAT term	AAT hierarchy	Notes
x10402	bone	animal	300011798	bone (material)	Materials Facet/ Materials (Hierarchy Name)/ <materials by composition>/ <combination inorganic/organic material>/ <combination inorganic/organic animal material> ALT: Objects Facet/ Components (Hierarchy Name)/ components (objects)/ <components by specific context> / <animal or human components>	Qualifier in label
x10369	bird claw	claw	n/a	n/a		There is "claw" but no "bird claw"
x10589	claw = talon	animal	300379351	claws	Objects Facet/ Components (Hierarchy Name)/ components (objects)/ <components by specific context> / <animal or human components>	NO parent in Materials Facet (compare to "bone"). Shows that hierarchies in different thesauri are more brittle/arbitrary than labels
n/a	n/a	n/a	300040710	talons	Objects Facet/ Components (Hierarchy Name)/ components (objects)/ <components by specific context>/ <furnishings components>/ <furniture components>/ <feet and foot components: furniture>/ <foot components>	In AAT "talons" has altLabel "claws" but "claws" doesn't have altLabel "talons" (which IMHO is an omission). In BM "claw" and "talon" are synonyms. In wordnet "claw" is a hypernym of "talon" (which is defined as a "sharp hooked claw"). The AAT case shows that two synonyms could be used in totally unrelated subtrees, so disambiguation is needed
x10370	bird egg	egg	n/a	n/a		There is "emu egg" but no "bird egg"
x10760	egg	animal product	300011801	egg	Materials Facet/ Materials (Hierarchy Name)/ materials (matter)/ <materials by origin>/ <biological material>/ animal material/ <animal material by form or function>/ <egg and egg components>	Perfect match
x10371	bird eggshell	eggshell	n/a	n/a		There is "ostrich eggshell" but no "bird eggshell"
x10761	eggshell	animal product	300191893	eggshell (animal material)	Materials Facet/ Materials (Hierarchy Name)/ materials (matter)/ <materials by origin>/ <biological material>/ animal material/ <animal material by form or function>/ <egg and egg components>/ <egg components>	Multiword qualifier in name. Another concept with different qualifier "eggshell (pottery style)" is unrelated. Shows importance of handling qualifiers (not just throwing them out) for disambiguation
x10372	bird of paradise feather	feather	n/a	n/a		No concept "bird of paradise" in AAT
x10373	bird of paradise skin	bird skin	n/a	n/a		No concept "bird of paradise" in AAT
x10375	bird skin	skin	300379982	bird skin	Materials Facet/ Materials (Hierarchy Name)/ materials (matter)/ <materials by origin>/ <biological material>/ animal material/ <collagenous material>/ skin (collagenous material)	Perfect match
x10374	bird of paradise tissue	bird tissue	n/a	n/a		No concept "bird of paradise" in AAT
x10377	bison bone	mammal bone	n/a	n/a	AAT has "bone (material)", and separately "Bison (genus)" in "Agents Facet/ Living Organisms/ Eukaryota (domain)/ Animalia (kingdom)/ Chordata (phylum)/ Vertebrata (subphylum)/ Mammalia (class)/ Artiodactyla (order)/ Bovidae (family)" BM: this is an example of a precoordinated (AND) term, which is generally considered a bad practice. BM: it's a bit unclear whether this means Bison (genus) or one of its two species: "Bison bison" (American bison) or "Bison bonasus" (European bison)"
x10378	bison hoof	hoof	n/a	n/a		See above
x10379	bison horn	horn	n/a	n/a		See above
x10380	bison skin	mammal skin	n/a	n/a		See above
x10381	bison wool	wool	n/a	n/a		See above
x11991	wool = felt	hair; textile	300243430	wool (textile)	Materials Facet/ Materials (Hierarchy Name)/ materials (matter)/ <materials by form>/ <materials by physical form>/ <fiber and fiber products>/ <fiber products>/ <textile materials>/ <textile materials by composition or origin>
			300014074	wool (hair)	Materials Facet/ Materials (Hierarchy Name)/ materials (matter)/ <materials by origin>/ <biological material>/ animal material/ <keratinous material>/ <hair and hair components>/ hair/ <hair by form>	BM term has two equally good exactMatches AAT says in related concepts: "object(s) using/produced by wool (hair) are wool (textile)". So AAT differentiates raw material from end product (which itself is a material), while BM does not
x10383	black bear skin	bear skin
x10384	black lily root	lily root
x11112	lily root	root