24 Apr
2006
24 Apr
'06
6:18 p.m.
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced innumerable automated
document classification systems. Many of these systems, such as those based on
the well-known Naive Bayes algorithm, treat each word as a distinct entity,
ignoring complex interactions between them. While for some applications this
approach may appear reasonable, the precise arrangements of words in political
documents often convey meanings which cannot be captured so easily. In this
paper, we investigate the success of such naive algorithms by comparing
Wordscores, a Naive Bayes derivative, to several well-known algorithms and a
new classification system based on vanilla recursive heirarchical
Dirichlet-multinomial mixture models, pointing out avenues for future
advancement. Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it shows dramatically increased performance, comparable to
some of the latest developments in document classification, at carrying out a
small number of carefully selected classifications on meticuously arranged
collections of political documents, and discuss its use in practical
applications.
24 Apr
24 Apr
6:40 p.m.
Dear Geoff & Chris,
I like your beginning a lot - very clear - but I would suggest breaking your
final sentence into 2 and perhaps getting rid of the final clause of the
second to last sentence, to maintain your clarity throughout.
Best,
Abby
Surprisingly, we find that the assumptions of Wordscores notwithstanding, it
shows dramatically increased performance at carrying out a small number of
carefully selected classifications on meticulously arranged collections of
political documents, comparable to some of the latest developments in
document classification. We conclude by discussing Wordscores' use in
practical applications.
-----Original Message-----
From: gov2001-l-bounces(a)lists.fas.harvard.edu
[mailto:gov2001-l-bounces@lists.fas.harvard.edu]On Behalf Of
ghumphr(a)fas.harvard.edu
Sent: Monday, April 24, 2006 10:19 PM
To: gov2001-l(a)lists.fas.harvard.edu
Subject: [gov2001-l] Preliminary Abstract
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced innumerable
automated
document classification systems. Many of these systems, such as those based
on
the well-known Naive Bayes algorithm, treat each word as a distinct entity,
ignoring complex interactions between them. While for some applications this
approach may appear reasonable, the precise arrangements of words in
political
documents often convey meanings which cannot be captured so easily. In this
paper, we investigate the success of such naive algorithms by comparing
Wordscores, a Naive Bayes derivative, to several well-known algorithms and a
new classification system based on vanilla recursive heirarchical
Dirichlet-multinomial mixture models, pointing out avenues for future
advancement. Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it shows dramatically increased performance, comparable to
some of the latest developments in document classification, at carrying out
a
small number of carefully selected classifications on meticuously arranged
collections of political documents, and discuss its use in practical
applications.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
6:54 p.m.
Geoff and Chris,
Good abstract. I would second Abby's point and, in addition,
suggest you say a bit more about the substance of your results. What
kind of political documents are you using to demonstrate the
usefulness of the process? What specifically are the practical
implications? Also, the title seems very "vanilla", which is ironic
given the vanilla recursive hierarchical Dirichlet-multinomial mixture
model. I have left some possible ideas below. Goodluck.
Sheldon
Exploring the Hidden Transcript - Political Documents and Structures of Meaning
Political Grammar: How Meaning is Constructed in Political Documents
On 4/24/06, Abby Williamson <abby.williamson(a)gmail.com> wrote:
Dear Geoff & Chris,
I like your beginning a lot - very clear - but I would suggest breaking your
final sentence into 2 and perhaps getting rid of the final clause of the
second to last sentence, to maintain your clarity throughout.
Best,
Abby
Surprisingly, we find that the assumptions of Wordscores notwithstanding, it
shows dramatically increased performance at carrying out a small number of
carefully selected classifications on meticulously arranged collections of
political documents, comparable to some of the latest developments in
document classification. We conclude by discussing Wordscores' use in
practical applications.
-----Original Message-----
From: gov2001-l-bounces(a)lists.fas.harvard.edu
[mailto:gov2001-l-bounces@lists.fas.harvard.edu]On Behalf Of
ghumphr(a)fas.harvard.edu
Sent: Monday, April 24, 2006 10:19 PM
To: gov2001-l(a)lists.fas.harvard.edu
Subject: [gov2001-l] Preliminary Abstract
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced innumerable
automated
document classification systems. Many of these systems, such as those based
on
the well-known Naive Bayes algorithm, treat each word as a distinct entity,
ignoring complex interactions between them. While for some applications this
approach may appear reasonable, the precise arrangements of words in
political
documents often convey meanings which cannot be captured so easily. In this
paper, we investigate the success of such naive algorithms by comparing
Wordscores, a Naive Bayes derivative, to several well-known algorithms and a
new classification system based on vanilla recursive heirarchical
Dirichlet-multinomial mixture models, pointing out avenues for future
advancement. Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it shows dramatically increased performance, comparable to
some of the latest developments in document classification, at carrying out
a
small number of carefully selected classifications on meticuously arranged
collections of political documents, and discuss its use in practical
applications.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
8:38 p.m.
Geoff and Chris:
I have only a minor point. I initially interpreted "Classifying
Political Documents" as a reference to the manner in which secret
government information is disclosed (i.e. "a classified memo was
circulated to the press"). I'm guessing you meant it as a pun (which
I like), but you might want to clarify the meaning with some sort of
subtitle so as to avoid confusion.
Omar
On Apr 24, 2006, at 10:54 PM, Sheldon Bond wrote:
Geoff and Chris,
Good abstract. I would second Abby's point and, in addition,
suggest you say a bit more about the substance of your results. What
kind of political documents are you using to demonstrate the
usefulness of the process? What specifically are the practical
implications? Also, the title seems very "vanilla", which is ironic
given the vanilla recursive hierarchical Dirichlet-multinomial mixture
model. I have left some possible ideas below. Goodluck.
Sheldon
Exploring the Hidden Transcript - Political Documents and Structures
of Meaning
Political Grammar: How Meaning is Constructed in Political Documents
On 4/24/06, Abby Williamson <abby.williamson(a)gmail.com> wrote:
Dear Geoff & Chris,
I like your beginning a lot - very clear - but I would suggest
breaking your
final sentence into 2 and perhaps getting rid of the final clause
of the
second to last sentence, to maintain your clarity throughout.
Best,
Abby
Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it
shows dramatically increased performance at carrying out a small
number of
carefully selected classifications on meticulously arranged
collections of
political documents, comparable to some of the latest developments in
document classification. We conclude by discussing Wordscores' use in
practical applications.
-----Original Message-----
From: gov2001-l-bounces(a)lists.fas.harvard.edu
[mailto:gov2001-l-bounces@lists.fas.harvard.edu]On Behalf Of
ghumphr(a)fas.harvard.edu
Sent: Monday, April 24, 2006 10:19 PM
To: gov2001-l(a)lists.fas.harvard.edu
Subject: [gov2001-l] Preliminary Abstract
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced innumerable
automated
document classification systems. Many of these systems, such as
those based
on
the well-known Naive Bayes algorithm, treat each word as a distinct
entity,
ignoring complex interactions between them. While for some
applications this
approach may appear reasonable, the precise arrangements of words in
political
documents often convey meanings which cannot be captured so easily.
In this
paper, we investigate the success of such naive algorithms by
comparing
Wordscores, a Naive Bayes derivative, to several well-known
algorithms and a
new classification system based on vanilla recursive heirarchical
Dirichlet-multinomial mixture models, pointing out avenues for future
advancement. Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it shows dramatically increased performance,
comparable to
some of the latest developments in document classification, at
carrying out
a
small number of carefully selected classifications on meticuously
arranged
collections of political documents, and discuss its use in practical
applications.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
Omar Wasow
M.A.-Ph.D. Candidate
Department of African and African American Studies
Department of Government
Harvard University
7:22 p.m.
I agree with the comments from my colleagues.
I think you explain clearly what the naive bayes approach does. I
wonder though if it is possible to explain in a sentence what a
"vanilla recursive heirarchical Dirichlet-multinomial mixture model"
does or is. Or how the other "well known algorithms" you utilize
differ from it.
Thomas
On Apr 24, 2006, at 10:18 PM, ghumphr(a)fas.harvard.edu wrote:
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced
innumerable automated
document classification systems. Many of these systems, such as
those based on
the well-known Naive Bayes algorithm, treat each word as a distinct
entity,
ignoring complex interactions between them. While for some
applications this
approach may appear reasonable, the precise arrangements of words
in political
documents often convey meanings which cannot be captured so easily.
In this
paper, we investigate the success of such naive algorithms by
comparing
Wordscores, a Naive Bayes derivative, to several well-known
algorithms and a
new classification system based on vanilla recursive heirarchical
Dirichlet-multinomial mixture models, pointing out avenues for future
advancement. Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it shows dramatically increased performance,
comparable to
some of the latest developments in document classification, at
carrying out a
small number of carefully selected classifications on meticuously
arranged
collections of political documents, and discuss its use in practical
applications.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
8:44 p.m.
That really is a good suggestion; I will have to give it some more thought.
I hope that all is well with you.
Geoff
Quoting Thomas Soehl <thomas_soehl(a)ksg07.harvard.edu>du>:
I agree with the comments from my colleagues.
I think you explain clearly what the naive bayes approach does. I
wonder though if it is possible to explain in a sentence what a
"vanilla recursive heirarchical Dirichlet-multinomial mixture model"
does or is. Or how the other "well known algorithms" you utilize
differ from it.
Thomas
On Apr 24, 2006, at 10:18 PM, ghumphr(a)fas.harvard.edu wrote:
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced
innumerable automated
document classification systems. Many of these systems, such as
those based on
the well-known Naive Bayes algorithm, treat each word as a distinct
entity,
ignoring complex interactions between them. While for some
applications this
approach may appear reasonable, the precise arrangements of words
in political
documents often convey meanings which cannot be captured so easily.
In this
paper, we investigate the success of such naive algorithms by
comparing
Wordscores, a Naive Bayes derivative, to several well-known
algorithms and a
new classification system based on vanilla recursive heirarchical
Dirichlet-multinomial mixture models, pointing out avenues for future
advancement. Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it shows dramatically increased performance,
comparable to
some of the latest developments in document classification, at
carrying out a
small number of carefully selected classifications on meticuously
arranged
collections of political documents, and discuss its use in practical
applications.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
7:24 p.m.
this sounds good. i would go a little bit farther in explaining some of
the terms to political scientists who don't know what naive bayes anything
is.
Gary
On Mon, 24 Apr 2006, ghumphr(a)fas.harvard.edu wrote:
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced innumerable automated
document classification systems. Many of these systems, such as those based on
the well-known Naive Bayes algorithm, treat each word as a distinct entity,
ignoring complex interactions between them. While for some applications this
approach may appear reasonable, the precise arrangements of words in political
documents often convey meanings which cannot be captured so easily. In this
paper, we investigate the success of such naive algorithms by comparing
Wordscores, a Naive Bayes derivative, to several well-known algorithms and a
new classification system based on vanilla recursive heirarchical
Dirichlet-multinomial mixture models, pointing out avenues for future
advancement. Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it shows dramatically increased performance, comparable to
some of the latest developments in document classification, at carrying out a
small number of carefully selected classifications on meticuously arranged
collections of political documents, and discuss its use in practical
applications.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
8:41 p.m.
In recent years, political methodologists, have produced innumerable automated
document classification systems. Many of these ignore word sequence
information, treating entire documents as mere collections of words. A subset
of these, including those based on the well-known Naive Bayes algorithm, assume
that word frequencies are independent. In this paper, we investigate the success
of such algorithms by comparing Wordscores, a Naive Bayes derivative, to several
well-known algorithms and a new classification system based on vanilla recursive
heirarchical Dirichlet-multinomial mixture models, pointing out avenues for
future advancement. Surprisingly, we find that its assumptions notwithstanding,
Wordscores shows dramatically increased performance, comparable to some of the
latest developments in document classification, at carrying out a small number
of carefully selected classifications on meticuously arranged collections of
political documents, and discuss its use in practical applications.
Quoting Gary King <king(a)harvard.edu>du>:
this sounds good. i would go a little bit farther in explaining some of
the terms to political scientists who don't know what naive bayes anything
is.
Gary
On Mon, 24 Apr 2006, ghumphr(a)fas.harvard.edu wrote:
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced innumerable
automated
document classification systems. Many of these
systems, such as those based
on
the well-known Naive Bayes algorithm, treat each
word as a distinct entity,
ignoring complex interactions between them. While for some applications
this
approach may appear reasonable, the precise
arrangements of words in
political
documents often convey meanings which cannot be
captured so easily. In this
paper, we investigate the success of such naive algorithms by comparing
Wordscores, a Naive Bayes derivative, to several well-known algorithms and
a
new classification system based on vanilla
recursive heirarchical
Dirichlet-multinomial mixture models, pointing out avenues for future
advancement. Surprisingly, we find that the assumptions of Wordscores
notwithstanding, it shows dramatically increased performance, comparable to
some of the latest developments in document classification, at carrying out
a
small number of carefully selected
classifications on meticuously arranged
collections of political documents, and discuss its use in practical
applications.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
8:55 p.m.
Actually, nm, I need to make some more changes.
I hope that all is well with you.
Geoff
Quoting ghumphr(a)fas.harvard.edu:
In recent years, political methodologists, have produced innumerable
automated
document classification systems. Many of these ignore word sequence
information, treating entire documents as mere collections of words. A subset
of these, including those based on the well-known Naive Bayes algorithm,
assume
that word frequencies are independent. In this paper, we investigate the
success
of such algorithms by comparing Wordscores, a Naive Bayes derivative, to
several
well-known algorithms and a new classification system based on vanilla
recursive
heirarchical Dirichlet-multinomial mixture models, pointing out avenues for
future advancement. Surprisingly, we find that its assumptions
notwithstanding,
Wordscores shows dramatically increased performance, comparable to some of
the
latest developments in document classification, at carrying out a small
number
of carefully selected classifications on meticuously arranged collections of
political documents, and discuss its use in practical applications.
Quoting Gary King <king(a)harvard.edu>du>:
this
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
this sounds good. i would go a little bit farther in explaining some of
the terms to political scientists who don't know what naive bayes anything
is.
Gary
On Mon, 24 Apr 2006, ghumphr(a)fas.harvard.edu wrote:
based
Geoff Humphreys and Chris Long
Classfying Political Documents
In recent years, political methodologists, have produced innumerable
automated
> document classification systems. Many of these systems, such as those on
> the well-known Naive Bayes algorithm, treat each word as a distinct
entity,
ignoring
complex interactions between them. While for some applications
this
approach may appear reasonable, the precise
arrangements of words in
political
> documents often convey meanings which cannot be captured so easily. In > paper, we investigate the success of such
naive algorithms by comparing
> Wordscores, a Naive Bayes derivative, to several well-known algorithms
and
a
> new classification system based on vanilla recursive heirarchical
> Dirichlet-multinomial mixture models, pointing out avenues for future
> advancement. Surprisingly, we find that the assumptions of Wordscores
> notwithstanding, it shows dramatically increased performance, comparable
to
> some of the latest developments in document
classification, at carrying
out
a
> small number of carefully selected classifications on meticuously
arranged
collections of political documents, and discuss its use in practical
applications.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
28 Apr
28 Apr
6:36 a.m.
New subject: [gov2001-l] Preliminary Abstract "Spillover Effects on Health Outcomes: Evidence from the Randomized Evaluation of PROGRESA"
Hi all,
Comments welcome.
Thanks,
Sebastian and Holger
Spillover Effects on Health Outcomes: Evidence from the Randomized
Evaluation of PROGRESA
Holger Lutz Kern and Sebastian Bauhoff
Abstract: We estimate spillover effects on health outcomes of PROGRESA, a
large-scale poverty alleviation program in Mexico. PROGRESA provides cash
incentives to eligible households if they participate in health education
and maternal and child health care. Exploiting the fact that the program
was assigned randomly across villages we find evidence for positive
spillover effects on non-eligible children aged 0-5 for objective and
subjective health measures. One year after the program began in 1998,
non-eligible children in treatment villages were on average [XX] taller,
[XX] heavier and [XX] per cent less likely to be anemic relative to
non-eligible children in control villages. They were also [XX] per cent less
likely to be reported sick. Previous evaluations have ignored these
spillover effects and hence understate the program's benefits.
11:15 a.m.
New subject: [gov2001-l] Preliminary Abstract "Spillover Effects on Health Outcomes: Evidence from the Randomized Evaluation of PROGRESA"
This is a very nice abstract. Your substantive point is clear and
important. Depending on your audience, and certainly for our purposes,
you might highlight the methodological contribution a bit more
clearly--precisely what methods are you bringing to bear on this problem?
Also, the title could be more of a statement of the argument, rather than
a statement of the topic.
Best,
Dan
On Fri, 28 Apr 2006, Sebastian Bauhoff wrote:
Hi all,
Comments welcome.
Thanks,
Sebastian and Holger
Spillover Effects on Health Outcomes: Evidence from the Randomized
Evaluation of PROGRESA
Holger Lutz Kern and Sebastian Bauhoff
Abstract: We estimate spillover effects on health outcomes of PROGRESA, a
large-scale poverty alleviation program in Mexico. PROGRESA provides cash
incentives to eligible households if they participate in health education
and maternal and child health care. Exploiting the fact that the program
was assigned randomly across villages we find evidence for positive
spillover effects on non-eligible children aged 0-5 for objective and
subjective health measures. One year after the program began in 1998,
non-eligible children in treatment villages were on average [XX] taller,
[XX] heavier and [XX] per cent less likely to be anemic relative to
non-eligible children in control villages. They were also [XX] per cent less
likely to be reported sick. Previous evaluations have ignored these
spillover effects and hence understate the program's benefits.
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
6576
days inactive
6580
days old
gov2001@lists.gking.harvard.edu
10 comments
8 participants
participants (8)
-
abby.williamson＠gmail.com -
bauhoff＠fas.harvard.edu -
dhopkins＠fas.harvard.edu -
ghumphr＠fas.harvard.edu -
king＠harvard.edu -
owasow＠fas.harvard.edu -
sheldon.bond＠gmail.com -
thomas_soehl＠ksg07.harvard.edu