Bart,
You are apparently aware of the main problem:
>
> Because ontologies may, and will define the same concepts differently
> (depending on the context, a subject matter expert's knowledge, their
> modelling approach, etc.), checking for consistency will be necessary,
> but also tricky, which I realize is a serious understatement. Most
> systems currently in place have some level of manual verification, but
> if a system's domain is to expand to a large repository, with an
> active community contributing often, a fully automated system would be
> desirable. This can naturally be extended to the web, and semantic
> classification of documents. My task now is to see how far an
> algorithm can go before some level of partial consistency is
> introduced, and how this introduction is executed.
>
. . . but perhaps you have still underestimated the magnitude. As one who
has tried to map ontologies to each other, I have concluded that it is
essentially impossible to do it accurately without participation of the
developers of the individual ontologies, because
(1) it requires not only human-level intelligence to understand the
meanings of the ontology components to be mapped, it requires an expert in
ontology to understand them; if we ever get to the point where a machine can
do this, automated ontology mapping will not be needed - the machines will
generate ontologies directly from text.
(2) even if a machine had that capability, it still wouldn't be able to
map accurately, because the documentation is never adequate to resolve
ambiguities, and almost always very sparse. Except in the simplest cases,
one needs the original developer to explain the intended meanings. (01)
So, if you understand that the mappings will always be of low quality
compared to a human product, and have an application in mind (such as
search) that can tolerate large errors, then good luck. (02)
But one final point - many of the mapping efforts I have seen seem to be
predicated on the notion that it will be impossible to get broad agreement
on some foundation ontology. I have been suggesting that this is quite
feasible, if there were enough funding to support an effort of adequate size
- at least 100 participants, to represent a broad range of user needs and
develop a foundation ontology that can serve as the basis for creating more
specialized ontologies. This would require a few tens of millions of
dollars and a process that emphasizes open-source data and accessory
utilities. No such project has ever been funded, and until it is I think
one should hesitate to opine that broad agreement will in fact be
impossible. I think that more than that amount has already been spent
studying mapping techniques, which are in my view an inadequate substitute.
I think that semi-automated mapping *will* be useful to link separately
generated domain ontologies to a common foundation ontology, once such a
common foundation ontology gains widespread use. (03)
But there may also be useful things that can be done with automatically
generated mappings, and if one must go that route I suppose that it is
worthwhile trying to improve the process. Just be aware that you are
tackling a project that is daunting for human experts, and be sure to know
what applications could use a result with a high error rate. (04)
Pat (05)
Patrick Cassidy
MICRA, Inc.
908-561-3416
cell: 908-565-4053
cassidy@xxxxxxxxx (06)
> -----Original Message-----
> From: ontolog-forum-bounces@xxxxxxxxxxxxxxxx [mailto:ontolog-forum-
> bounces@xxxxxxxxxxxxxxxx] On Behalf Of Bart Gajderowicz
> Sent: Tuesday, March 24, 2009 1:50 AM
> To: ontolog-forum@xxxxxxxxxxxxxxxx
> Subject: [ontolog-forum] Automated Ontology Mapping
>
> Hello,
>
> This is my first posting on this forum, so let me introduce myself.
> My name is Bart Gajderowicz, and I'm a graduate student at the
> Computer Science Department at Ryerson University, in Toronto, Canada.
>
> I am researching automated ontology mapping, and have compiled several
> options which span the different categories currently being developed.
> Let me formally define these as per (Choi, et al 2006). Based on this
> work, I am currently looking at the several fields/ideas to introduce
> partial consistency to automate the mapping process.
>
> I welcome any comments, corrections, suggestions, criticisms from the
> forum on the following analysis.
>
> Some of the current techniques take a deductive approach, and
> concentrate on the structures, axioms, and hierarchies of the
> ontology. Others take an inductive approach, and look at instances in
> order to derive what objects are being modelled. Others still are a
> hybrid of the two. At this point I'm looking at both approaches, to
> see which direction is more appropriate for my research.
>
> I'm currently concentrating my efforts on ontologies defined by first
> order languages such as Common Logic. I'm doing this to take
> advantage of provers and inference engines, but also to limit my
> domain to these languages, so compatible representation becomes less
> of a headache.
>
> At this point, I would also like to concentrate on structural and
> taxonomic similarities, with a limited amount of lexical similarity
> measures. Preferably, no natural language processing would be
> performed on terms at this time. This may limit my ability to perform
> schema or semantic mapping. In FOL, however, I have the ability to
> apply unification techniques on a set of axioms, to align free and
> bound variables.
>
> Because ontologies may, and will define the same concepts differently
> (depending on the context, a subject matter expert's knowledge, their
> modelling approach, etc.), checking for consistency will be necessary,
> but also tricky, which I realize is a serious understatement. Most
> systems currently in place have some level of manual verification, but
> if a system's domain is to expand to a large repository, with an
> active community contributing often, a fully automated system would be
> desirable. This can naturally be extended to the web, and semantic
> classification of documents. My task now is to see how far an
> algorithm can go before some level of partial consistency is
> introduced, and how this introduction is executed.
>
> To that end, I am currently looking at the following fields/ideas to
> introduce partial consistency:
>
> 1) A similarity measure can be evaluated by looking at properties such
> as isomorphism, injection, surjection, associativity, commutativity,
> and distributivity between some classes, but not others, in different
> ontologies.
>
> 2) Perhaps Prototype Theory will allow me to formally define some key
> terms, as a type of local upper ontology. In OntoClean (Guarino, et
> al 2004), for example, the first process was to define a "backbone
> taxonomy of terms" to which all other terms were grounded.
>
> 3) Fuzzy Logic may also be suitable, by creating membership functions
> which include some sets of axioms attributes, or properties, but not
> others, and overlapping relations may be true to some degree. As was
> pointed out to me, these would define a maximum entropy, which is used
> in the data mining field, not logic. This may still be worth looking
> at, for inductive analysis. Perhaps a redefinition of inconsistent
> FOL clauses to Fuzzy Logic could be done in a such a way that intended
> meaning of axioms are persevered, and theories are weakened to
> accommodate the discrepancies. This approach would greatly limit the
> decidability of theories, but if key axioms are pointed out, with
> corresponding degrees, perhaps this limitation is worth the gains.
> GLUE, for example, uses a Naive Bayes learner to analyse statements
> of ontologies.
>
> 4) Perhaps adding Modal Logic into the mix, and saying that some set
> of properties define a concept now, but not at a later time, could be
> a way of quantifying differences between concepts in different
> ontologies. Think of a Wikipedia entry where the definition of a term
> may change with popular opinion or when more information becomes
> available.
>
> 5) Contexts may be generated where some axioms are added or removed.
>
>
> References:
>
> Choi, N., Song, I.-Y., Han, H., A survey on ontology mapping (2006),
> SIGMOD Record, 35 (3), pp. 34-41.
>
> Guarino, N., Welty, C., Evaluating ontological decisions with
> ontoclean (2002), Communications of the ACM, 45 (2), pp. 61-65.
>
> Thank you.
> --
> Bart Gajderowicz
> MSc Candidate, '10
> Dept. of Computer Science
> Ryerson University
> http://www.scs.ryerson.ca/~bgajdero
>
> _________________________________________________________________
> Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
> Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
> Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
> Shared Files: http://ontolog.cim3.net/file/
> Community Wiki: http://ontolog.cim3.net/wiki/
> To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
> To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx
> (07)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (08)
|