Ferenc and John,
Thanks for the call for papers, though India is a little too far for
comfortable travel given the present state of aircraft terminals. I do
agree that what pattern recognition technology has long called features, or
invariant value sets, should be given more shrift in database technology, and finding
such patterns might be eased by a proper ontology interpreter which has been
fit specifically to the application.
As I understand the Tree of Porphyry, "Things" are the
starting point at the root, and predefined questions can be asked at each node about
predefined properties at that node to traverse from the top node (root: all
things) to the actually specified thing ("Sally Jennings, teacher, daughter
of Peter Jennings") at the terminal node when there are no more questions
to ask.
The porphyric approach is very different than starting with scoped sets
of properties, as Ferenc suggested. In the tree of Porphyry, John, you
are starting with the set of all things (at the root) and asking set narrowing questions
about specified properties and their values re members of the set until you
reach an end point in the tree with only zero or one member thing - a hierarchy
by construction, and predicates at each node are based on fluents, by
construction. See 7,209,923 for details. (auto URL formation short
circuited below to avoid spam cops):
Englishlogickernel.com/Patent-7-209-923-B1.pdf
That approach works great to distinguish among TYPES that have already
been defined through FCA. I am more interested in the huge clutter of
textual information in databases that has not already been typed, parsed, fit
into a framed meaning, and "interpreted". I think that material
is more complex than people realize, and can be very valuable.
The FCA method is to cluster groups of things that all have the
"same" properties, and to divide the members into subsets or elements
based on fluents, or on constraints among the properties. So anything
visual (at least those visualized by humans) must have properties such as color, intensity, shape, shade, etc. That
is as true for a book as for a salamander. But it doesn’t relate
equilateral triangles to equiangular ones, unless the designer throws that in
too. So Ferenc’s suggestion has nothing DIRECTLY to do with TYPE
construction per se, but with an inventory of exactly that set of properties
that can be identified by the observer at each node in the porphyric node in
current question.
This is also a different approach from starting with Things, as in the
poryphyral approach.
Another way to look at it is to begin with a set of "primitive
properties" (John, you will enjoy that word) which can be observed.
Only after defining what those observables are, can you construct the
tree of porphyry and specifically identify all the TYPES of things that can
possibly have values for each of those properties. To begin with, there
are no questions for the things at the root node of the tree. So you have
to know what to measure about Thing existence, how to represent those
measurements, and how to compare two such measurements.
Starting with a set of defined properties limits the distinctions that
can be made to manage the debate/discourse about how to represent Things. THAT
is a property, IMHO.
HTH,
-Rich
Sincerely,
Rich Cooper
EnglishLogicKernel.com
Rich AT EnglishLogicKernel DOT com
-----Original Message-----
From: ontolog-forum-bounces@xxxxxxxxxxxxxxxx
[mailto:ontolog-forum-bounces@xxxxxxxxxxxxxxxx] On Behalf Of John F. Sowa
Sent: Sunday, February 07, 2010 9:20 AM
To: [ontolog-forum]
Subject: Re: [ontolog-forum] foundation ontology primitives
Ferenc and Rich,
FK> The current ontologies focus on objects only, properties are not
> structured into repositories and relations are now defined
in a
> funny idiosyncratic way that does not allow integration of
ontologies.
> The contained in, or partof relation, etc. too are very
trivial and
> uninteresting from a NL translation point of view.
I agree that much richer approaches are needed. Properties are
important *monadic* predicates. The ContainedIn and PartOf are just
two among the many, many important dyadic predicates. The
thematic
roles or case relations are other important linguistic relations.
But just as important, if not more so, are the structural operators
for combining larger, prefabricated patterns. The Framenet group
at Berkeley
has been working on that:
http://framenet.icsi.berkeley.edu/
Chuck Fillmore, who was one of the first to emphasize the importance
of case relations in his famous paper "The Case for Case",
was also
the prime mover in starting the Framenet project.
The approach that I use with conceptual graphs was inspired by some
of the work on case relations combined with the semantic networks
of AI. Following is a review article in the Encyclopedia of AI:
http://www.jfsowa.com/pubs/semnet.htm
Semantic Networks
For conceptual graphs, I defined six "canonical formation
rules" for
combining graph structures. My original version of 1976 contained
four rules, the 1984 version contained a different selection of four,
and the latest version (since 2000) has six rules, organizes as three
pairs of generalization/specialization rules. See Section 3 of
http://www.jfsowa.com/cg/cg_hbook.pdf
Conceptual Graphs
RC> Your hint about structuring properties into repositories is an
> interesting one. Other than the very simple clustering
approach
> of FCA (formal concept analysis), I don’t know of
approaches that
> START with properties, other than my own stuff
> http://www.englishlogickernel.com
> and partition properties into groups.
Actually, Aristotle thought of it first. The Tree of Porphyry
from
the 3rd century AD (see Figure 1 in the semnet.htm article) showed
how categories and properties are interrelated. Ramon Lull in the
13th century had rotating disks for combining properties, and
Leibniz was inspired by Lull to develop the first computational
approach to developing a lattice of concept types. For a brief
summary of both Leibniz's Universal Characteristic and FCA, see
http://www.jfsowa.com/logic/math.htm#Lattice
There have also been various conferences about the section of features
or properties for data mining. See, for example, the following
CFP.
John Sowa
=============================================================================
International Workshop on Feature Selection in
Data Mining (FSDM10)
21st of June 2010, Hyderabad,
India
(In conjunction with PAKDD 2010)
http://featureselection.asu.edu/fsdm10
Knowledge discovery and data mining (KDD) is a multidisciplinary effort
to mine gold nuggets of knowledge from data. The increasingly large
data
sets from many application domains have posed unprecedented challenges
to KDD; in the meantime, new types of data are evolving such as social
media, text, and microarray data, to name a few. Researchers and
practitioners in multiple disciplines and various IT sectors confront
similar issues in feature selection, and there is a pressing need for
continued exchange and discussion of challenges and ideas, exploring
new
methodologies and innovative approaches to generate breakthroughs.
Feature selection is effective in data preprocessing and reduction that
is an essential step in successful data mining applications. Feature
selection has been a research topic with practical significance in many
areas such as statistics, pattern recognition, machine learning, and
data mining (including Web, text, image, and microarrays). The
objectives of feature selection include: building simpler and more
comprehensible models, improving data mining performance, and helping
prepare, clean, and understand data. Workshop on Feature Selection in
Data Mining (FSDM2010) aims to further the cross-discipline,
collaborative effort in variable and feature selection research.
FSDM2010 will be held at the 14th Pacific-Asia Conference on Knowledge
Discovery and Data Mining (PAKDD 2010)
The workshop invites all papers related to feature selection, and
especially welcomes contributions that highlight emerging feature
selection challenges in data mining. Possible paper topics include, but
are not limited to:
- Dimensionality reduction
- Feature weighting
- Feature ranking
- Subset selection
- Feature extraction/construction
- Feature selection methodology
- Integration with data mining algorithms
- Pitfalls and learned lessons in feature selection studies
- Novel data structures
- Selection in small sample domains
- Data streams and time series
- Feature selection bias and variance
- Selection in extremely high-dimensional domains
- Real-world case studies and applications that highlight the role of
feature selection
- Emerging challenges
--- KEY DATES ---
Paper Submission deadline: March 19th, 2010
Author Notification: April 16th, 2010
Camera-ready: April 30th, 2010
Workshop: June 21th, 2010
--- ORGANIZATION ---
Huan Liu, Hiroshi Motoda, Rudy Setiono, Zheng Zhao
_________________________________________________________________
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