Re: [Dbpedia-discussion] type of http://dbpedia.org/page/Bachelor_of_Arts

Hi guys,

the problem of DBpedia entity types is not a new one. I have discussed some of the problems in the paper [1]. 

In my opinion it is pretty hard to establish the correct type of an entity using only one source of information. DBpedia uses Infoboxes (both infobox types and recently it performs type inference based on individual properties), YAGO uses categories while Tipalo uses first sentences. But each method has its inherent difficulties.
In our recent research we try to combine these methods in order to provide most accurate typing of DBpedia entites. The intermediate results are available at [2]. Still they are not perfect, but we are making progress. E.g. we do not have "Bachelor of Arts", but we have: "Bachelor of Arts in Applied Psychology" and similar, which are classified as EducationalDegree. We use Cyc and Umbel concepts as types of the entities. 

If anyone is interested in that research, I am eager to discuss it during the forthcoming ISWC in Italy.

Cheers,
Aleksander

[1] A. Pohl, Classifying the Wikipedia Articles into the OpenCyc Taxonomy [in:] Proceedings of the Web of Linked Entities Workshop in conjuction with the 11th International Semantic Web Conference, Giuseppe Rizzo, Pablo Mendes, Eric Charton, Sebastian Hellmann, Aditya Kalyanpurs (eds.), p. 5-16, ISSN: 1613-0073. 

[2] http://klon.wzks.uj.edu.pl/wiki-types/


---- Wł. Pn, 13 paź 2014 17:51:07 +0200 Heiko Paulheim<heiko@informatik.uni-mannheim.de> napisał(a) ---- 


  Hi Valentina,
 
 I am not sure whether I understand you correctly. There might be cases of metonymy in DBpedia, but as far as I can see, Wikipedia is usually quite good at separating them via disambiguation pages, I am not sure whether there are too many example.
 
 The problem with the degrees, as far as I can tell, is not a metonymy one (degrees are just degrees, I have never seen them used to refer to a university), but simply a series of shortcomings in DBpedia. What happens here inside DBpedia is the following:
 * First, we find an infobox which says that someone's almaMater is, say, "Princeton University (B.A.)". Both Princeton and B.A. are linked to the respective Wikipedia pages.
 * The extraction framework extracts two statements from that: 
 PersonX almaMater Princeton_University, and
 PersonX almaMater Bachelor_of_Arts
 (the second one being an error, which is very hard to avoid in the general case)
 * Since that happens a few times, we infer that Bachelor_of_Arts is a University.
 
 So in that case, I think it's purely a DBpedia problem. If you are aware of any actual cases of metonymy, however, I am curious to hear about that.
 
 All the best,
 Heiko
 
 
 
 Am 13.10.2014 16:33, schrieb Valentina Presutti:
 
 38-F25A-4A56-857A-1A31400D20C4@gmail.com" type="cite"> Hi Heiko, 
 
 thanks for the prompt reply and the explanation.
 However, the interesting thing is that these entities are clearly used with more than one sense (at least in the US culture), so the issue comes from this fact originally in my opinion.
 I mentioned two cases here, but if you check you can see that all these types of entities (Degrees) have the same problem.
 
 
 My suggestion (if that can help) is to identify such metonym cases and have a special approach: having different entities as the number of senses.
 
 
 However, the Wikipedia page of such entities defines them as degrees…not sure if this can be useful to notice for you. 
 
 
 Valentina
 
  On 13 Oct 2014, at 09:03, Heiko Paulheim <heiko@informatik.uni-mannheim.de> wrote:
 
  Hi Valentina,
 
 (and CCing the DBpedia discussion list)
 
 this is an effect of the heuristic typing we employ in DBpedia [1]. It works correctly in many cases, and sometimes it fails - as for these examples (the classic tradeoff between coverage and precision). 
 
 To briefly explain how the error comes into existence: we look at the distribution of types that occur for the ingoing properties of an untyped instance. For dbpedia:Bachelor_of_Arts, there are, among others, 208 ingoing properties with the predicate dbpedia-owl:almaMater (which is already questionable). For that predicate, 87.6% of the objects are of type dbpedia-owl:University. So we have a strong pattern, with many supporting statements, and we conclude that dbpedia:Bachelor_of_Arts is a university. That mechanism, as I said, works reasonable well, but sometimes fails at single instances, like this one. For dbpedia:Academic_degree, you'll find similar questionable statements involving that instace, that mislead the heuristic typing algorithm.
 
 With the 2014 release, we further tried to reduce errors like these by filtering common nouns using WordNet before assigning types to instances, but both "Academic degree" and "Bachelor of Arts" escaped our nets here :-(
 
 The public DBpedia endpoint loads both the infobox based types and the heuristic types. If you need a "clean" version, I advise you to set up a local endpoint and load only the infobox based types into it.
 
 Best,
 Heiko
 
 [1] http://www.heikopaulheim.com/documents/iswc2013.pdf
 
 
 
 
 Am 13.10.2014 02:42, schrieb Valentina Presutti:
 
 BE-430E-4480-BA32-BB9B69EE3377@gmail.com" type="cite">Dear all, 
 
 I noticed that dbpedia:Bachelor_of_Arts, as well as other similar entities (dbpedia:Bachelor_of_Engineering, dbpedia:Bachelor_of_Science, etc.), is typed as dbpedia-owl:University
 I would expect a type like “Academic Degree” but if you look at
 dbpedia:Academic_Degree, its type is again dbpedia-owl:University
 
 
 however, its definition is (according to dbpedia):
 
 
 "An academic degree is a college or university diploma, often associated with a title and sometimes associated with an academic position, which is usually awarded in recognition of the recipient having either satisfactorily completed a prescribed course of study or having conducted a scholarly endeavour deemed worthy of his or her admission to the degree. The most common degrees awarded today are associate, bachelor's, master's, and doctoral degrees.”
 
 
 Showing that there are at least two different meanings associated with the term: college/university and title.
 I thing that different meanings should be separated so as to allow applications to refer to the different entities: a university or a title.
 
 
 At least for me this causes errors in automatic relation extraction...
 
 
 Wdyt?
 
 
 Valentina
  
 -- Prof. Dr. Heiko Paulheim Data and Web Science Group University of Mannheim Phone: +49 621 181 2646 B6, 26, Room C1.08 D-68159 Mannheim Mail: heiko@informatik.uni-mannheim.de Web: http://www.heikopaulheim.com/  
  
 
 
  
 -- Prof. Dr. Heiko Paulheim Data and Web Science Group University of Mannheim Phone: +49 621 181 2646 B6, 26, Room C1.08 D-68159 Mannheim Mail: heiko@informatik.uni-mannheim.de Web: www.heikopaulheim.com  ------------------------------------------------------------------------------ 
Meet PCI DSS 3.0 Compliance Requirements with EventLog Analyzer 
Achieve PCI DSS 3.0 Compliant Status with Out-of-the-box PCI DSS Reports 
Are you Audit-Ready for PCI DSS 3.0 Compliance? Download White paper 
Comply to PCI DSS 3.0 Requirement 10 and 11.5 with EventLog Analyzer 
http://p.sf.net/sfu/Zoho_______________________________________________ 
Dbpedia-discussion mailing list 
Dbpedia-discussion@lists.sourceforge.net 
https://lists.sourceforge.net/lists/listinfo/dbpedia-discussion