Granularity in Library Linked Open Data
Fractals
- self-similar at all levels of granularity
- each circle represents of things that look very similar (snowflake looking pattern but of different sizes)
- characteristic of fractals
- cannot determine level: all levels are equal, some more equal than others
Multi-Faceted Granularity
- What is described by a bibliographic record? or a single statement?
- What is the level of description? How complete is it? e.g. AACR2
- How detailed is the schema used? How dumb? – especially relevant right now. The more detailed, the higher level of granularity possible.
- Semantic constraints? Unconstrained?
Resource Description Framework – Linked Data
- Triple: This resource | has intended audience | Juvenile
- Subject / Predicated / Object
- do each of these parts have granularity?
- higher/lower level, but should talk about coarse or fine grained granularity
Subject: What is the Statement About?
- we can focus on description an article / resource / work, then think about coarser or finer granularity:
- coarser: consortium collection / RDF map
- library collection / digital collection
- super-aggregate journal title / jurnal index
- aggregate: issue / festschrift
- focus on description an article / resource / work
- component: section / graphics / page
- sub-component: paragraph / markup
- finer: word rdf/xml
- uri / node
Predicate: What is the Aspect Described?
- similar coarse/fine breakdown:
- membership category
- access to resource
- access to content
- suitability rating
- audience and usage
- audience
- audience of audio-visual material
- diagram: possible audience map (partial) – unconstrained version to avoid collisions of isbd/dct/schema/rda/m21/frbrer
- different links can be made while still retain proper semantic links
- currently constructing just one giant graph
What is the Aspect Described?
- coarse to fine:
- resource record
- manifestation record
- title and s.o.r
- title statement
- title of manifestation
- title word
- first word of title
- why do librarians need so many titles? Why not just use dublin core title and be done with it? Because we need it to do our work e.g. spine title to browse
- title = string identifier
- RDA: what to do with this? how do we apply these needs?
- possible semantic map (partial) – I won’t even try to reproduce this
- need to take into account names and ranges
- make it more difficult, but more powerful
Semantic Reasoning: The Sub-Property Ladder
- this is where the graph becomes useful and property
- machines can’t reason, so we’re demantic the semantics such that we can give the rules to machines to process our data
- semantic rule:
- if property1 sub-property of property2;
- then data triple: resource property1 “string”
- implies data triple: resource property2 “string”
- otherwise, data triple remains the same
- simple enough for computer to carry out
- doesn’t matter how complex the map actually is, because it can still do it in matters of seconds
- machine entailment: isbd” “hast title proper” (finer) -> dct: “has title” (coarser)
- might sound simple, but making a computer do interferance
- ‘dumb(ing)-up, data has been lost, but still meaningful – moved from one schema to another
Data Triples from Multiple Schema / Entailed from Sub-Property Map / rom Property Domains
- frbrer: “has intended audience” – “primary school”
- isbd: “has note on use or audience” – “for ages 5-9”
- rda: “intended audience (work)” – “for children aged 7-“
- m21: “target audience” 0> m21terms: -> “Juvenile”
- definition attached to the vocabulary
- also talking about granularity
- can map the sub-property to top level of unc: “has note on use or audience”
- “is a” frbrer: “work”, isbd: “resource”, rda: “work” – rda and frbr schema actually separate, not semantically linked even though vocabulary is similar and RDA is based on FRBR
- once stabalized can be drawn from each other
What is the Aspect Described?
- coarser to finer:
- creator
- author
- screenwriting
- animation screenwriting
- children’s cartoon screenwriting
- different controlled vocabulary
- graph of RDA for author/creator/screenwriting in relation to work and agent
- graph of same thing, but for dc for creator and agent
- what is the semantic relationship between the dct creator and the rda creator?
- marcrel author maps to dc contributor, not creator – what is the relationship between rda author and marcrel author?
- decision from 2005, needs to be reappraised and reviewed
- relationship between dc creator and dc contributor?
- how does lcsh “screenwriters” fit?
Machine-Generated Granularity
- also has issues
- e.g. full-text indexing: down to the word level
- BabelNet: A very large multilingual ontology
- can get quite complex and granular
User-Generated Granularity
- users can actually generate useful metadata
- can use statistical methods to remove extremes and come back with consensus
- going to cause granularity problems e.g. “OK for my kids (7 and 9)”, “Too childish for me (age 14)”
KISS
- keep it simple, stupid
- keep it simple and stupid?
- data model is very simple: triples!
- in terms of complexity, actually very simple
- but metadata content is complex
- and therefore, resource discovery is complex
- complex structure of application of simple rules, similar in the hard sciences and math
- simplicity is elegance
AAA
- Anyone can say anything about any thing
- someone will say something about every thing
- in every conceivable way
- and then constrained linguistically
OWA
- open world assumption: the absence of a statement is not a statement of non-existence
Will it get so granular that it becomes too complex?
And the rest is science
Break Time
