SIMS 290-2: Applied Natural Language Processing: Marti Hearst

Download Report

Transcript SIMS 290-2: Applied Natural Language Processing: Marti Hearst 

Text categorization
Feature selection: chi square test
1
Joint Probability Distribution
The joint probability distribution for a set of random variables X1…Xn
gives the probability of every combination of values
P(X1,...,Xn)
Sneeze
Cold
0.08
¬Cold
0.01
¬Sneeze
0.01
0.9
The probability of all possible cases can be calculated by summing
the appropriate subset of values from the joint distribution.
All conditional probabilities can therefore also be calculated
P(Cold | ¬Sneeze)
BUT it’s often very hard to obtain all the probabilities for a joint
distribution
Slides adapted from Mary Ellen Califf
2
Bayes Independence Example
Imagine there are diagnoses ALLERGY, COLD, and WELL and
symptoms SNEEZE, COUGH, and FEVER
Can these be correct numbers?
Prob
P(d)
P(sneeze|d)
P(cough | d)
Well
0.9
0.1
0.1
Cold
0.05
0.9
0.8
Allergy
0.05
0.9
0.7
P(fever | d)
0.01
0.7
0.4
Slides adapted from Mary Ellen Califf
3
KL divergence (relative entropy)
P( x)
D( P || Q)   P( x) log
Q( x)
x
Basis of comparing two probability distributions
4
Text Categorization Applications
Web pages organized into category hierarchies
Journal articles indexed by subject categories (e.g., the Library
of Congress, MEDLINE, etc.)
Responses to Census Bureau occupations
Patents archived using International Patent Classification
Patient records coded using international insurance categories
E-mail message filtering
News events tracked and filtered by topics
Spam
Slide adapted from Paul Bennet
5
Yahoo
News
Categories
6
Text Topic categorization
Topic categorization: classify the document into
semantics topics
The U.S. swept into the Davis
Cup final on Saturday when twins
Bob and Mike Bryan defeated
Belarus's Max Mirnyi and Vladimir
Voltchkov to give the Americans
an unsurmountable 3-0 lead in the
best-of-five semi-final tie.
One of the strangest, most
relentless hurricane seasons on
record reached new bizarre heights
yesterday as the plodding approach
of Hurricane Jeanne prompted
evacuation orders for hundreds of
thousands of Floridians and high
wind warnings that stretched 350
miles from the swamp towns south
of Miami to the historic city of St.
Augustine.
7
The Reuters collection
A gold standard
Collection of (21,578) newswire documents.
For research purposes: a standard text collection to compare
systems and algorithms
135 valid topics categories
8
Reuters
Top topics in Reuters
9
Reuters Document Example
<REUTERS TOPICS="YES" LEWISSPLIT="TRAIN" CGISPLIT="TRAINING-SET"
OLDID="12981" NEWID="798">
<DATE> 2-MAR-1987 16:51:43.42</DATE>
<TOPICS><D>livestock</D><D>hog</D></TOPICS>
<TITLE>AMERICAN PORK CONGRESS KICKS OFF TOMORROW</TITLE>
<DATELINE>
kicks off
CHICAGO, March 2 - </DATELINE><BODY>The American Pork Congress
tomorrow, March 3, in Indianapolis with 160 of the nations pork producers from 44 member states
determining industry positions on a number of issues, according to the National Pork Producers
Council, NPPC.
Delegates to the three day Congress will be considering 26 resolutions concerning various issues,
including the future direction of farm policy and the tax law as it applies to the agriculture sector.
The delegates will also debate whether to endorse concepts of a national PRV (pseudorabies virus)
control and eradication program, the NPPC said.
A large trade show, in conjunction with the congress, will feature the latest in technology in all
areas of the industry, the NPPC added. Reuter
&#3;</BODY></TEXT></REUTERS>
10
Classification vs. Clustering
Classification assumes labeled data: we know how
many classes there are and we have examples for
each class (labeled data).
Classification is supervised
In Clustering we don’t have labeled data; we just
assume that there is a natural division in the data
and we may not know how many divisions (clusters)
there are
Clustering is unsupervised
11
Categories (Labels, Classes)
Labeling data
2 problems:
Decide the possible classes (which ones, how many)
Domain and application dependent
Label text
Difficult, time consuming, inconsistency between
annotators
12
Reuters Example, revisited
Why not topic = policy ?
<REUTERS TOPICS="YES" LEWISSPLIT="TRAIN" CGISPLIT="TRAINING-SET"
OLDID="12981" NEWID="798">
<DATE> 2-MAR-1987 16:51:43.42</DATE>
<TOPICS><D>livestock</D><D>hog</D></TOPICS>
<TITLE>AMERICAN PORK CONGRESS KICKS OFF TOMORROW</TITLE>
<DATELINE>
kicks off
CHICAGO, March 2 - </DATELINE><BODY>The American Pork Congress
tomorrow, March 3, in Indianapolis with 160 of the nations pork producers from 44 member states
determining industry positions on a number of issues, according to the National Pork Producers
Council, NPPC.
Delegates to the three day Congress will be considering 26 resolutions concerning various issues,
including the future direction of farm policy and the tax law as it applies to the agriculture sector.
The delegates will also debate whether to endorse concepts of a national PRV (pseudorabies virus)
control and eradication program, the NPPC said.
A large trade show, in conjunction with the congress, will feature the latest in technology in all
areas of the industry, the NPPC added. Reuter
&#3;</BODY></TEXT></REUTERS>
13
Binary vs. multi-way classification
Binary classification: two classes
Multi-way classification: more than two classes
Sometime it can be convenient to treat a multi-way
problem like a binary one: one class versus all the
others, for all classes
14
Features
>>> text = "Seven-time Formula One champion
Michael Schumacher took on the Shanghai circuit
Saturday in qualifying for the first Chinese
Grand Prix."
>>> label = “sport”
>>> labeled_text = LabeledText(text, label)
Here the classification takes as input the
whole string
What’s the problem with that?
What are the features that could be useful for
this example?
15
Feature terminology
Feature: An aspect of the text that is relevant to the
task
Some typical features
Words present in text
Frequency of words
Capitalization
Are there NE?
WordNet
Others?
16
Feature terminology
Feature: An aspect of the text that is relevant to the
task
Feature value: the realization of the feature in the text
Words present in text
Frequency of word
Are there dates? Yes/no
Are there PERSONS? Yes/no
Are there ORGANIZATIONS? Yes/no
WordNet: Holonyms (China is part of Asia),
Synonyms(China, People's Republic of China, mainland China)
17
Feature Types
Boolean (or Binary) Features
Features that generate boolean (binary) values.
Boolean features are the simplest and the most
common type of feature.
f1(text) = 1
0
f2(text) = 1
0
if text contain “elections”
otherwise
if text contain PERSON
otherwise
18
Feature Types
Integer Features
Features that generate integer values.
Integer features can be used to give classifiers access to
more precise information about the text.
f1(text) = Number of times “elections” occurs
f2(text) = Number of times PERSON occurs
19
2 statistic (CHI)
2 statistic (pronounced “kai square”)
A commonly used method of comparing proportions.
Measures the lack of independence between a term and
a category
20
2 statistic (CHI)
Is “jaguar” a good predictor for the “auto” class?
Term = jaguar
Term  jaguar
Class = auto
2
500
Class  auto
3
9500
We want to compare:
the observed distribution above; and
null hypothesis: that jaguar and auto are independent
21
2 statistic (CHI)
Under the null hypothesis: (jaguar and auto independent):
How many co-occurrences of jaguar and auto do we expect?
If independent: Pr(j,a) = Pr(j)  Pr(a)
So, there would be N  Pr(j,a), i.e. N  Pr(j)  Pr(a)
occurances of “jaguar”
Pr(j) = (2+3)/N;
Pr(a) = (2+500)/N;
N=2+3+500+9500
N(5/N)(502/N)=2510/N=2510/10005  0.25
Term = jaguar
Term  jaguar
Class = auto
2
500
Class  auto
3
9500
22
2 statistic (CHI)
Under the null hypothesis: (jaguar and auto independent):
How many co-occurrences of jaguar and auto do we expect?
Term = jaguar
Term  jaguar
expected: fe
Class = auto
Class  auto
2 (0.25)
3
500
9500
observed: fo
23
2 statistic (CHI)
Under the null hypothesis: (jaguar and auto – independent):
How many co-occurrences of jaguar and auto do we expect?
Term = jaguar
Term  jaguar
expected: fe
Class = auto
Class  auto
2 (0.25)
3 (4.75)
500
(502)
9500 (9498)
observed: fo
24
2 statistic (CHI)
2 is interested in (fo – fe)2/fe summed over all table entries:
 2 ( j , a)   (O  E ) 2 / E  (2  .25) 2 / .25  (3  4.75) 2 / 4.75
 (500  502) 2 / 502  (9500  9498) 2 / 9498  12.9 ( p  .001)
The null hypothesis is rejected with confidence .999,
since 12.9 > 10.83 (the value for .999 confidence).
Term = jaguar
Term  jaguar
expected: fe
Class = auto
Class  auto
2 (0.25)
3 (4.75)
500
(502)
9500 (9498)
observed: fo
25
2 statistic (CHI)
There is a simpler formula for 2:
A = #(t,c)
C = #(¬t,c)
B = #(t,¬c)
D = #(¬t, ¬c)
N=A+B+C+D
26