home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.ai:4683 sci.math.stat:2664
- Newsgroups: comp.ai,sci.math.stat
- Path: sparky!uunet!gatech!usenet.ins.cwru.edu!agate!boulder!ucsu!yertle.Colorado.EDU!mcclella
- From: mcclella@yertle.Colorado.EDU (Gary McClelland)
- Subject: Re: Learning from subjective data
- Message-ID: <mcclella.725048643@yertle.Colorado.EDU>
- Sender: news@ucsu.Colorado.EDU (USENET News System)
- Nntp-Posting-Host: yertle.colorado.edu
- Organization: University of Colorado, Boulder
- References: <BzE5G3.Hoq@ux1.cso.uiuc.edu> <ALMOND.92Dec21220007@bass.statsci.com> <Bzo3Fr.7Kw@cs.uiuc.edu>
- Date: Tue, 22 Dec 1992 18:24:03 GMT
- Lines: 38
-
- bharat@cs.uiuc.edu (R. Bharat Rao) writes:
-
-
- >almond@statsci.com (Russell G. Almond) writes:
- >>R. Bharat Rao (bharat@cs.uiuc.edu) writes:
- >>> I was wondering if anyone knew of any work that has been done on
- >>> learning from subjective data. For instance, you may have a data set
- >>> of events with a number of independent attribute (x1...xn) and a
- >>> single dependent attribute y. However, y is a subjective rating.
-
- >>This is generally a messy problem and I don't know that there has ever
- >>been a definative answer. I would, however, try the Psych--Stat
- >>literature,
-
- I second that recommendation. You may be particularly interested
- in some research on expert judgment by James Shanteau. He has
- studied, among other things, the subjective judgments of
- livestock judges and soil experts in agricultural contexts. This
- is the same generic problem as your esthetic judgment problem.
- Also of interest would be Ken Hammond's work. A useful starting
- place would be an aritcle in Science by Hammond & Adelman in the
- mid 70's.
-
-
- >OK, the case where you have a unique expert for every problem is
- >obviously very messy. Does the situation become any simpler if each
- >expert does many ratings (again no point is rated by more than one
- >expert)? Say, you have 10,000 data points and 500 experts each rate
- >20 random points. Also, for a point you know the expert who rated
- >that point. I think this makes the problem somewhat easier.
-
- Sure, just include terms in your model for the experts. This
- controls for differences among experts so long as the experts
- still agree as to which end is up for each attribute.
-
- gary mcclelland
- univ of colorado
- mcclella@yertle.colorado.edu
-