Re: [URW3 public] OWL extensions [was Re: [URW3] ... three questions based on the last telecon]

Yes Umberto, this form of specification we need. Of course there are variants of particular tools in the chain and particular models because we have to work with the web as it is. We can also  suggest web creator models (assuming he/she is not human page/service oriented and there are no machine access protection measures (some pages prevent robot access)). Yet there are several alternatives and even more combinations and all of these have to be experimentally implemented and evaluated, more in text

----- Original Message -----
From: Umberto Straccia [mailto:umberto.straccia@isti.cnr.it]
To: public-xg-urw3@w3.org
Subject: Re: [URW3 public] OWL extensions [was Re: [URW3] ...  three questions  based on the last telecon]


> 
> Hi Peter,
> 
> thanks for the response. But, my question rises many more  
> representation/reasoning issues, some cover issues of your previous  
> email. More generally, to do it automatically
> 
> 1. the query has to be represented in some formal user local language

There are alternatives (an unskilled user) e.g. QBE like querying, letting user evaluate a sample of objects, allow NLP queries? ...

> 2. a user local system (agent?) should be able to identify a subset  
> of relevant resources (web services?) to which to submit the query  
> (this is called automated  resource selection in IR - Information  
> Retrieval)

Again there are several alternatives, e.g. use Google API to locate services (I do not go into web service languages, interface communication,...) 

> 3. Then, the local query has to be reformulated (using ontology  
> alignment mappings) for each identified resource and submitted

here a big variety can occur, a pure html page (mostly text and or with richer structure of html tags), an xhtml page, annotated with RDFa or some tagging (using either pure NLP expressions or some microformats)

Ontology alignement is a big problem, here is a big space for uncertainty modelling, question is when does it take place, in the time of source location and selection, or later. Where does it take place on local computer or the agent ...

> 4. each resource provides back a ranked list, which then has to be  
> merged (this is called rank aggregation + data fusion in the literature)

this is the ideal case (as I have wrote, we have a model for this with an B+ tree index sitting on the source side), I agree that we should not orient ourselves only to user side but also to sugest web source authors to enrich their resources and make it more machine understandable. I do not know how to achieve it without their cooperation

> 5. (optional) the data has then to be translated back into the user's  
> local vocabulary (again, using ontology alignment mappings)
> 

yes, also there is a possibility to provide user with links to best / top-k answers...

> 

I am happy about this discussion. Your description Umberto is exactly what I have in mind with use case "What is web specific versus problems that are hard even on your computer"

Greetings Peter

Received on Thursday, 26 July 2007 08:38:31 UTC