RDB2RDF Working Group Teleconference -- 27 Apr 2010

<trackbot> Date: 27 April 2010

<mhausenblas> scribenick: Marcelo

<juansequeda> dialing in 1 min

<mhausenblas> juansequeda++

Admin

<cygri> cygri is with juansequeda

<mhausenblas> PROPOSAL: accept minutes from http://www.w3.org/2010/04/20-rdb2rdf-minutes.html

RESOLUTION: accept minutes from http://www.w3.org/2010/04/20-rdb2rdf-minutes.html

Main topic for today: Use cases document

Discussion about the use case document

<mhausenblas> http://www.w3.org/2001/sw/rdb2rdf/use-cases/

Main changes from the previous version of the document:

requirements have been updated

goal for today: agree on the docuement

<soeren> +q

<souri> Mapping and Transforming are two different goals. RDB2RDF charter does not include "Transforming".

<mhausenblas> Marcelo?

Sorry, I lost the phone connection

<mhausenblas> oh

<souri> Transforming should be done as a separate task, not under RDB2RDF.

<mhausenblas> scribenick: nunolopes

<Marcelo> I am back ...

<Marcelo> yes

<mhausenblas> scribenick: Marcelo

<souri> Mapping from RDB schema to a (generated) RDF schema is what we have to handle. Mapping the generated RDF schema to another (existing) ontology is an ontology-mapping problem (e.g., PROMPT).

<mhausenblas> scribenick: nunolopes

<Marcelo> sorry, I lost the connection again, I don't know what is going on :-(

cygri: the interesting cases are maping the RDB to RDF

… about the putative ontology, i'm not sure what the job of this group would be

<mhausenblas> Michael: noting that we lost some of the earlier statements due to scribe difficulties

<mhausenblas> Michael: I encourage people (soeren, etc.) to repeat their statements on IRC

… if we do a mechanical transformation from one format to the other what is the job of this group?

… is there any need for this group to address the issue/usecase of the putative ontology?

<souri> step-1) RDB -> (generated) ontology, and step-2) (generated) ontology to a second ontology

MacTed: the desired end result is not to map to a putative ontology

<souri> step-2 is a different problem that people have already attempted (e.g., PROMPT) and should be considered as a separate task

Ahmed: I question the value of mapping the RDB schema to an ontology

<souri> if we want to include step-2 (i.e., ontology-to-ontology mapping) in our roadmap then we need to include that in the RDB2RDF charter

<souri> people, please respect the queue

MacTed: I feel that the mapping should be general

souri: I agree with MacTed

… One problem is translating a view of the RDBMS to an ontology

… and another step is to translate the ontology to another ontology

… this is a lot of work and non trivial, and probably not in the charter

Ahmed: I don't see a group to translate the schema to a "local ontology"

cygri: if I understand, you want a separation where first is an automatic transformation from RDB to RDF

… and the second step (from a ontology to another) is outside the scope of the group?

… this is a valid approach

… but we have number of mapping languages that do these things in one step

… and the advantage is that you can specify the mapping as SQL queries

… which would not be doable in the separated approach

scribe: having a combined mapping language, allows to use the expressivity of SQL

<souri> step-1 includes writing SQL queries to influence the RDF schema that will be generated

souri: the 1st step (schema to general ontology) you want to use SQL

… the mapping is influenced by the SQL you are writting

<MacTed> +1

… those queries are then used to create the RDF schema

cygri: we should map databased to a given RDF schema

MacTed: which is ontology to ontology mapping

cygri: I don't think so

souri: what you mean by ontology, is that it consists of the database tables and for that I write a view

… the view is then translated automatically to the ontology

… we shouldn't care about the specific tables but more on the actual view

… the mapping relies on looking at tables from the DB

… this should be our goal

Ahmed: if it is impossible to map to a domain ontology, your DB is not intersection with your application

… otherwise we cannot have any mapping

Dan: ontology mapping is an open issue: what happens when we create software to do it automatically

…. it's not an open issue if you provide the mapping

<souri> ontology-to-ontology mapping is a schema-mapping problem which is nontrivial

… the charter of the group is to create a mapping language, a syntax that says what is a mapping

… object X in a database goes to Y in the ontology

… by reading the charter I don't think that there is a lot of dispute

mhausenblas: I'm a bit lost. we planned for today to review the use case document

… we're falling back to the question of what we should do

<Ahmed> +Ahmed

… we either postpone the question or make available what is currently online

<Zakim> mhausenblas, you wanted to remind people on the goal: get out the UC document. today.

<souri> This is how I see it: A human sees the target ontology, the human sees the DB schema, and the human then figures out in his/her head what the SQL queries (if we use SQL-based approach) should be and writes those queries in SQL. The mapping specification includes the list of SQL queries and the desired RDF class names and properties.

Ahmed: I would not provide a document while people are still discussing what the group should do

… I don't see the mapping as mapping to a local ontology but for a domain ontology

<mhausenblas> strawman poll on the following topics:

MacTed: does RDF require a domain ontology ?

… or does the use of a putative ontology deliver RDF?

<souri> I'd like to have a comment first to explain the context around the question that will be put up for straw-poll

… we should opt for the generic solution

souri: we are figuring out if we are using SQL what will map to the RDF

<cygri> mhausenblas: MacTed's position: It is sufficient for this group to specify a mapping from database to a putative ontology. The result from this is RDF, and that satisfies the goal of this group. Going from this simplest possible RDF to a domain ontology is out of scope.

…. we are using the mapping we define by using SQL

Dan: to clarify local ontology vs domain ontology

<hhalpin> Note that the charter is a bit vague about "The mapping language MUST define the mapping of relational data and relational schemas to RDF and OWL."

<hhalpin> whether a domain or putative ontology is used.

… as souri said there is a mapping to "a" ontology

<hhalpin> I am not sure if domain or putative ontology are the words the database community uses.

<Zakim> juansequeda, you wanted to ask what if somebody wants to map to FOAF or SIOC or GoodRelations... etc

… either an existing ontology or something just made up

juansequeda: if we're not mapping to a domain ontology, what do you do yo people who expect your data to be mapped to a specific ontology

MacTed: that's the next step

<hhalpin> However, I would say transformation from a straightforward mapping is *within* scope, and whether this is to be done with SQL views, SPARQL, or some other way is to be decided on some level as an option in R2ML..

… use backward-chaining, not forcing the triples to be materialised

<cygri> mhausenblas: My position: For this group to produce anything useful, it has to provide an answer to the problem of getting from database data to FOAF, GoodRelations etc

… there are cases where the SQL query directly maps to an ontology

… but that is a limited subset of the use cases

hhalpin: this is a usecase and requirement document

… we should no specifiy what we will be using

… the language should allow straightforward mapping

<mhausenblas> please no more people on the queue now. Ahmed is last, then we do the poll

…. or allow to map more

… we can delay if we don't reach consencus

<hhalpin> And remember that we're not doing the technical work in the use-case and requirement document :)

cygri: regarding souri's comments, I agree with the general workflow

… it's a matter of how to do it

<hhalpin> We have to specify it re use-cases and expressivity, no more.

<hhalpin> The question to me seems to be over expressivity, i.e. should we have *any*.

… at this point we should not say anything that assumes it should be done by creating views or some other way

<souri> <DB> ===> (try to fit it to your target ontology by writing SQL queries, mapping of SQL queries to classes, query-result-cols to RDF properties) ===> virtual RDF data based on target on target ontology

<hhalpin> I would not concur with MacTed, i.e. we should allow (a perhaps optional) expressivity beyond a simple direct mapping.

… we should focus on what we say in the requirements document

<hhalpin> we should not build any technical solution into the use-case document.

<mhausenblas> The two options for the strawman poll read:

souri: the mapping should be figured by the human

<mhausenblas> 1. It is sufficient for this group to specify a mapping from database to a putative ontology. The result from this is RDF, and that satisfies the goal of this group. Going from this simplest possible RDF to a domain ontology is out of scope.

… the question is what I have to do on top of the database schema to make it look like the target ontology

<mhausenblas> 2. For this group to produce anything useful, it has to provide an answer to the problem of getting from database data to FOAF, GoodRelations etc, we need to do both steps

… we prefer the SQL approach because we are oracle

… but if you don't want to use it it's fine

<hhalpin> I agree, Souri's approach is not out of scope.

<hhalpin> In fact, I am arguing that it is in scope.

<hhalpin> But we don't want to specify it on the technical level, just the expressivity level.

<souri> why don't we postpone the voting on it instead of rushing to it

straw poll

<hhalpin> happy to put it into a WBS poll

<hhalpin> however, I will not get to it till this evening.

<hhalpin> I am at WWW2010.

<MacTed> <-- 1

<souri> the choices have to be understood well

<juansequeda> 2

<mhausenblas> 2

<Ahmed> 2

<hhalpin> 2

<Marcelo> 2

<Dan> 2

<souri> could you please put the choices here first

<hhalpin> 1. Just putative

<Ashok> 2

<souri> I don't remember the choices yet

<hhalpin> 2. Putative and domain ontology

<whalb> 2

<hhalpin> That was the options!

<lma2> 2

<hhalpin> Perhaps a poll is not necessary.

<soeren> 2

<souri> 0 -- the choices are not clear to me

<hhalpin> Let's chat over e-mail with MacTed to see how we can phase the distinction to make sure that it addresses your concerns, i.e. we should not *force* people to match to a domain ontology (i.e. to transforms besides direct mapping in *some* language)

- DRAFT -

RDB2RDF Working Group Teleconference

27 Apr 2010

Attendees

Contents

Admin

Discussion about the use case document

straw poll

Summary of Action Items

Scribe.perl diagnostic output