Paul Houle wrote: > Kingsley Idehen wrote: >> >> One approach, using SIOC ontology: >> Try C as a property of an Entity of Type: sioc:Container that is >> associated with another Entity of Type: sioc:Item that has properties >> for x (Place) and y (Person). De-referencing the sioc:Container URI >> will expose data for C, and via the "sioc:container_of" property you >> will get to data for x and/or y. >> > I've been thinking a lot about this over the weekend. > > I think the really valuable thing I could provide isn't the > matrix, but a set of facts that are derived from the matrix but that > are filtered by human effort. That's fine also. > > The "A Newspaper is an Organization" fact is a good example. It > can be easily stated in RDFS, and live along side the assertion that > "A Newspaper is a Creative Work". In a detailed model, we could > distinguish between > > (1) The New York Times (as a bibliographic entry) New York Times Newspaper. As per: http://dbpedia.org/resource/The_New_York_Times > (2) The New York Times Corporation (which also owns the Boston Globe, > WQXR-FM, 15 other newspapers and 15% of the Boston Red Sox) As per: http://dbpedia.org/resource/The_New_York_Times_Company > (3) The Business Unit inside the NYTC which produces the New York Times. > > If we follow the principle that "dbpedia is about wikipedia", then > I think we can say that > > http://en.wikipedia.org/wiki/The_New_York_Times > > is about (1) and (3), and that
> > http://en.wikipedia.org/wiki/The_New_York_Times_Company > > is about (2). So overall, I think it's right to say that > > http://en.wikipedia.org/wiki/The_New_York_Times > > is both a Work and an Organization. The Work and the Organization > are certainly conflated in the "commonsense" model that most people have. > > Thinking about it over the weekend, I've realized that I'm free to > assert any triples I want. I can publish my own additions to the > dbpedia ontology, and people are free to use them or not use them. > The one thing that I shouldn't do is mint new URI's under dbpedia's > namespace: if I did want to add new types or predicates to the > dbpedia ontology, I need to do them in my own namespace. Yes, and want you state is a little bigger than you can imagine. Basically, a major Linked Data Web "blind spot" is the notion that we loose the ability to hold and express our opinions in triples hosted in our own Linked Data Spaces :-) We can say what we want since there aren't any absolute truths. The key thing is to expose our personal views via our personal linked data spaces. Believe it or not, you are touching on the very principle behind: OpenLink Data Spaces (ODS). The single most misunderstood piece of Linked Data technology I know :-) > > But then another batch of questions comes up. > > In the desktop publishing age, an individual could publish a > newspaper by themselves. However, it wouldn't be much of a > newspaper, not notable enough to be in Wikipedia unless it gets > involved in some bizzare controversy. I think that the fact "A > Newspaper is an Organization" does much more good than harm, but > reasoning systems that use such facts need some ability to deal with > uncertainty. (Default logic?) > > The "Person and Place are disjoint" assertion is similar, but > worse. In particular, there are about 10 counterexamples. > Considering how many Persons and Places there are, and the nature of > Wikipedia, that's excellent data quality. I've worked on line of > business systems that are a lot worse. What does one do with such an > assertion? Reject the whole data set? Reject the offending > triples? Paint the offending triples red? If reasoning is conditional and scoped to inference rules (associated with your data space) that are optionally applied to SPARQL queries etc. Thus, I am saying: put your axioms in some place, and they can be conditionally applied to SPARQL queries directed at your data space. > > I feel comfortable making assertions in SKOS and OWL when I let > concepts like 'disjoint', 'same as', and 'subclass' have the kind of > fuzzy meaning that these terms have for people. I don't feel so > comfortable making them with the semantics given by the standards and > actual implementations of the RDF/RDFS/OWL stack. > > The best I can see doing is to split up the assertions that I make > into several groups, putting the "safer" ones together and the more > "dangerous" ones together. That at least gives people some control > about what they're going to use. Pack the axioms into separate graph IRIs (so make separate data sets). This is what I've done with Kavitha's DBpedia to Freebase Type mappings. Links: 1. http://ods.openlinksw.com/wiki/ODS/ Kingsley > > > > > > -- Regards, Kingsley Idehen Weblog: http://www.openlinksw.com/blog/~kidehen President & CEO OpenLink Software Web: http://www.openlinksw.com ------------------------------------------------------------------------------ Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ Dbpedia-discussion mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/dbpedia-discussion
