lecture14-SVMs-handout-6-per

45

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: as
grain
   The
interes)ng
theore)cal
answer
is
to
explore
semi‐ supervised
training
methods:
   In
prac)ce,
rules
get
a
lot
bigger
than
this
   Can
also
be
phrased
using
~
or
~.idf
weights
   With
careful
crauing
(human
tuning
on
development
 data)
performance
is
high:
   Construe:
94%
recall,
84%
precision
over
675
categories
 (Hayes
and
Weinstein
1990)
   How
can
you
insert
yourself
into
a
process
where
humans
 will
be
willing
to
label
data
for
you??
   Amount
of
work
required
is
huge
   Es)mate
2
days
per
class
…
plus
maintenance
 Introduc)on to Informa)on Retrieval   Bootstrapping,
EM
over
unlabeled
documents,
…
   The
prac)cal
answer
is
to
get
more
labeled
data
as
 soon
as
you
can
 33
 Sec. 15.3.1 34
 Introduc)on to Informa)on Retrieval Sec. 15.3.1 A
reasonable
amount...
View Full Document

This document was uploaded on 02/26/2014.

Ask a homework question - tutors are online