Web Zealots

View the world's data in the SemWeb.

• unambiguous
  Err on the side of not re-using over not understanding.
• "semantic"
  Opt-in term reuse for e.g. foaf:givenName.
• answers useful questions

This talk: <http://w3.org/brief/MjQx>

(W3C) Use Cases

eGov

Publication/access to civil data.

Life Sciences

SPARQL query of decentralized biomedical data.

Health Care

Matching coded clinical protocols.

This talk: <http://w3.org/brief/MjQx>

Database to DLG

Database to DLG

Database to DLG

Database to DLG

Database to DLG

Direct Mapping

• n-ary primary keys
• n-ary foreign keys
• NULL-y
• decomposable row identifers
• decomposable column identifers
• minimally escaped
  <http://伝言.example/?employee=أكرم,task=R&D>

<People/ID=7> <People#ID> 7 .
<People/ID=7> <People#fname> "Bob" .
<People/ID=7> <People#addr> <Addresses/ID=18> .

<People/ID=8> <People#ID> 8 .
<People/ID=8> <People#fname> "Sue" .

<Addresses/ID=18> <Addresses#ID> 18 .
<Addresses/ID=18> <Addresses#city> "Cambridge" .
<Addresses/ID=18> <Addresses#state> "MA" .
      

Direct Graph

• + Exposes relational graph.
• ± Reflects SQL schema.
• − Doesn't use popular RDF schema;
  e.g. <People#fname> vs. foaf:name.
• − No control of graph structure.

<People/ID=7> <People#ID> 7 .

<People/ID=7> <People#fname> "Bob" .
<People/ID=7> <People#addr> <Addresses/ID=18> .
<People/ID=8> <People#ID> 8 .

<People/ID=8> <People#fname> "Sue" .

<Addresses/ID=18> <Addresses#ID> 18 .
<Addresses/ID=18> <Addresses#city> "Cambridge" .

<Addresses/ID=18> <Addresses#state> "MA" .
      

RDB2RDF Approaches

R2RML

D2R-like mapping declaration

Virtuoso RDF Views

DDL-like definition of an RDF graph

SquirrelRDF, Direct Mapping

   simple dump of data.
+ rules

Rules Languages

• RIF BLD
• SWRL
• RuleML
• SPARQL CONSTRUCT

The semantics align well, but implementations focus on different use case optimizations.

CONSTRUCT Usage Patterns

   PREFIX :mydb <http://cityhospital.example/dbs>
CONSTRUCT { ?o a               study:SubjectObservation .
            ?o study:subject   ?p .
            ?o study:clinician ?d .
            ?d :foaf:name ?dName }

    WHERE { ?o mydb:patient ?p .
            ?o mydb:doctor  ?d .
            ?d mydb:name    ?dName }

5-star Data

What makes data sharable?

• Same concepts
• in the same model
• using the same schema
• connecting the same identifiers

SPARQL1.1 strings and term constructors

• strings: substring, upper/lower case, string arithmetic
• terms: IRIs, blank nodes, datatyped and language-tagged strings
• Assignment, e.g. SELECT (f(?foo) AS ?bar) addresses some identifier divergence.

  PREFIX uni: <http://purl.uniprot.org/uniprot/>
  ?target <Protein#uniprot> ?ulabel .
  BIND (IRI(CONCAT(uni:, ?ulabel)) AS ?uniprotID)

Domain-appropriate Views

• In SQL, many views are virtual (and some are updatable).
• An SQL view definition looks like a rule:
  

  CREATE VIEW foo (
    SELECT genes.id AS gene, labels.text AS label
      FROM genes
      JOIN lables ON genes.label = lables.id 
  )

• An RDF view definition (ab)using SPARQL CONSTRUCTs.

  CONSTRUCT {
      ?gene uniprot:id ?id ; skos:prefLabel ?gene_symbol
    } WHERE {
      _:gene Ugene:acc ?id ; Ugene:val ?gene_symbol
    }

• Views upon views...

Query Transformation

SELECT ?o ?d
 WHERE {
   ?o study:subject <Bob> .
   <Bob> study:clinician ?d .
 }

SELECT ?o ?d
 WHERE {
   ?o mydb:onPatient <Bob> .
   ?o mydb:byDoctor ?d .
 }

SELECT observations... doctors...
  FROM observations, 
   patients ON id=observations.patient ...
   WHERE patients.name = "Bob"

SPARQL to SQL

How easy will it be to turn a query over the above graph
into an SQL query over the input tables?

SELECT ?name
 WHERE {
   ?who db:name "Bob" .
   ?who db:manager ?boss .
   ?boss db:name ?name .
 }

SELECT boss.name
FROM Employees AS who
INNER JOIN Employees AS boss ON boss.id=who.manager
WHERE who.name="Bob"

• join consistency
• constraints push-down
• cardinality

SPARQL to SPARQL

• General Query Transformation
  • Graph pattern match against rule heads.
  • Query body instantiated from variables and constants bound in head.
  • Query may entail multiple rule head instantiations.
    A foaf:know B. B foaf:knows C. requires 2 X ?x foaf:knows ?y
  • Unbound vars made unique.
• Complex query patterns handled recursively.
• Most commonly used to compose IRIs:
  BIND (IRI(CONCAT(uni:, ?ulabel)) AS ?uniprotID)

Function Inversion

Resolve as much as possible;
huge impact on performance.

{ <http://www.uniprot.org/uniprot/P04637> skos:prefLabel ?symbol }

=>

SELECT (fn:lower-case(?_uniProt_0_u_gene_symbol) AS ?symbol) 
 WHERE {
   _:_uniProt_0_gene <http://ucsc.example/uniProt/gene#acc> "P04637"  .
   _:_uniProt_0_gene <http://ucsc.example/uniProt/gene#val> ?_uniProt_0_u_gene_symbol .
}

=>

SELECT TOLOWER(uniProt.val) AS ?symbol
FROM uniProt WHERE acc = "P04637"

SELECT TOLOWER(uniProt.val) AS ?symbol
FROM uniProt WHERE CONCAT("http://www.uniprot.org/uniprot/", uniprot.acc) =
                          "http://www.uniprot.org/uniprot/P04637"

Function Inversion

Resolve as deeply as possible.

• String functions:
  CONCAT("a", ?b, "c", ?d) = "a123bxyz": → regexp("a(.*?)c(.*?)")
• Algebraics:
  ?a + ?b = 3
  • ?a bound → ?b bound to 3-?a
  • ?a bound → FILTER (?a + ?b = 3)
• Lossy functions:
  fn:upper-case(?s) = "ASDF"
  • optimistic → ?s bound "asdf"
  • pesimistic → FILTER (fn:upper-case(?s) = "ASDF")

Questions/Comments

• How (procedure)?
• Why (motivation)?
• Who (vendor participation)?
• When (RSN).
• Where (W3C's RDB2RDF WG).