lecture18-learning-ranking-handouts-6-per

1541 introducontoinformaonretrieval

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: 
how
do
we
 tell
how
good
our
scoring
func)on
is?
 Introduc)on to Informa)on Retrieval Sec.
6.1.2
 Introduc)on to Informa)on Retrieval Choosing
g
 Choosing
g
   In
our
simple
sesng,
all
that
magers
is
the
number
 of
examples*
of
each
equivalence
class
   Define:
   n01r
=
#
examples
with
sT=0,
sB=1,
judgment
=
Rel
   n01n
=
#
examples
with
sT=0,
sB=1,
judgment
=
NonRel
   n10r
=
#
examples
with
sT=1,
sB=0,
judgment
=
Rel
   n10n
=
#
examples
with
sT=1,
sB=0,
judgment
=
NonRel
   (and
similarly
n00r, n00n, n11r and n11n corresponding
to
 the
4
other
equivalence
classes)
   The
n01r
examples
with
sT=0,
sB=1
combined
 contribute
a
total
least‐squared
error
of
   Similarly,
add
up
the
error
contribu)ons
of
the
other
 3
combina)ons
of
sT
and
sB for
a
total
error
of
 * this may not hold for other sets of features, e.g., the # characters in the query 2 5/30/11 Introduc)on to Informa)on Retrieval Introduc)on to Informa)on Retrieval Choosing
g is
now
elementary
calculus
 Generalizing
this
simple
example
   Differen)a)ng
the
total
error
wrt
g
we
get
the
 op)mal
value
for
g
to
be
   More
(than
2)
features
   Non‐Boolean
fea...
View Full Document

This document was uploaded on 02/26/2014.

Ask a homework question - tutors are online