- From: Michael K. Bergman <mike@mkbergman.com>
- Date: Mon, 12 May 2008 17:14:09 -0500
- To: akshay bhat <akshayubhat@gmail.com>
- CC: dbpedia-discussion@lists.sourceforge.net, public-lod@w3.org
Hi Akshay, Chris Bizer wrote: > > Hi Akshay, > >> Conceptnet is a commonsense database in form of a large semantic >> network containing >> 150,000 concepts and 700,000 assertions. http://conceptnet.media.mit.edu/ >> The concepts are linked to each other by 23 predicates. ( eg. Dog isa >> pet ) >> currently i am trying to bring this dataset ( creative commons 3.0 >> license ) >> on semantic web by linking with dbpedia. > > The description of Conceptnet on the project page sounds like this > ontology could be very useful within many projects and it would be great > to have it on the Semantic Web! > >> However as i am new to this field i request your advice on how it can >> be done. >> Till now i have been successful in linking the good quality (more than >> one assertion) >> concepts from concpetnet with cwcc classes and yago classes. >> ( the concepts are linked with categories as all member of the >> categories can be considered as related to the concept) >> The data is available from here >> >> aubhat2.googlepages.com >> >> Before linking the concepts with wikipedia articles (as done in cyc >> dataset) >> i would like to get your opinions regarding the similarity metrics >> which can be used. > > Frederick and Mike (cc'ed) are currently doing similar work with > interlinking their UMBEL ontology http://www.umbel.org/ with DBpedia and > Yago. > > I guess they would be the right people to talk to about similarity metrics. We, too, looked closely at ConceptNet and were quite intrigued. To my knowledge, the system is not being as actively maintained as previously, and that was a concern to us. But, that was only one factor among many that caused us to select OpenCyc as our reference concept basis. [1] We welcome you to follow the UMBEL project via its Google group [2] and to look for the coming documentation of how we explicitly vetted and mapped from OpenCyc. That documentation in three-volumes, due out hopefully by the end of this month, may provide some ideas for how you might tackle a ConceptNet mapping. I would guess there is a full person-year of efforts in our UMBEL stuff to date, so don't be surprised if your own effort is quite a bit to chew! As for similarity metrics, Chris has commented elsewhere on this and we have some new UMBEL predicates that deal with similar but non-equivalent class-class alignments (isAligned), entity-class (isAbout) and entity-entity (isLike) with associated predicates for inverse properties and similarity metrics. (The UMBEL ontology documentation and reference files will also be released at the same time as the OpenCyc vetting volumes.) Applying any of these predicates may be pretty tricky. The use of similarity metrics ("confidence intervals" if you will), especially, is a new area, and as I have written elsewhere, possibly as likely as nuclear waste to be widely embraced. :). In the interim, you may want to look at the Dec 2007 Yago technical report [3] and its Web site [4] for one approach. You may also want to look at some of the recent SKOS discussion on these issues of similarity alignments.[5] Good luck! Thanks, Mike [1] http://www.mkbergman.com/?p=433 [2] http://groups.google.com/group/umbel-ontology [3] http://www.mpi-inf.mpg.de/~suchanek/publications/yagotr.pdf [4] http://www.mpi-inf.mpg.de/~suchanek/downloads/yago/ [5] http://www.w3.org/2008/05/06-swd-minutes.html, see SKOS Mapping section at end > > For information about publishing your data, you could have a look at > http://www4.wiwiss.fu-berlin.de/bizer/pub/LinkedDataTutorial/ > > Keep on the good work! > > Cheers > > Chris > > >> >> -- >> akshay uday bhat. >> department of chemical engineering >> university institute of chemical technology >> mumbai India > > _______________________________________________ > Dbpedia-discussion mailing list > Dbpedia-discussion@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/dbpedia-discussion > > -- __________________________________________ Michael K. Bergman CEO Zitgist LLC 319.621.5225 skype:michaelkbergman http://zitgist.com http://mkbergman.com __________________________________________
Received on Monday, 12 May 2008 22:14:58 UTC