Text Classification

and Naïve Bayes

The Task of Text

Classification
Dan Jurafsky

Is this spam?
Dan Jurafsky

Who wrote which Federalist papers?

• 1787-8: anonymous essays try to convince New York

to ratify U.S Constitution: Jay, Madison, Hamilton.
• Authorship of 12 of the letters in dispute
• 1963: solved by Mosteller and Wallace using
Bayesian methods

James Madison Alexander Hamilton

Dan Jurafsky

Positive or negative movie review?

• unbelievably disappointing
• Full of zany characters and richly applied satire, and some
great plot twists
• this is the greatest screwball comedy ever filmed
• It was pathetic. The worst part about it was the boxing
scenes.

4
Dan Jurafsky

What is the subject of this article?

MEDLINE Article MeSH Subject Category Hierarchy

• Antogonists and Inhibitors
• Blood Supply
• Chemistry
? • Drug Therapy
• Embryology
• Epidemiology
• …
5
Dan Jurafsky

Text Classification
• Assigning subject categories, topics, or genres
• Spam detection
• Authorship identification
• Age/gender identification
• Language Identification
• Sentiment analysis
• …
Dan Jurafsky

Text Classification: definition

• Input:
• a document d
• a fixed set of classes C = {c1, c2,…, cJ}

• Output: a predicted class c Î C

Dan Jurafsky
Classification Methods:
Hand-coded rules
• Rules based on combinations of words or other features
• spam: black-list-address OR (“dollars” AND“have been selected”)
• Accuracy can be high
• If rules carefully refined by expert
• But building and maintaining these rules is expensive
Dan Jurafsky

Classification Methods:
Supervised Machine Learning
• Input:
• a document d
• a fixed set of classes C = {c1, c2,…, cJ}
• A training set of m hand-labeled documents (d1,c1),....,(dm,cm)
• Output:
• a learned classifier γ:d à c

9
Dan Jurafsky
Classification Methods:
Supervised Machine Learning
• Any kind of classifier
• Naïve Bayes
• Logistic regression
• Support-vector machines
• k-Nearest Neighbors

• …
Text The Naive Bayes Classifier
Classification
and Naive
Bayes
Naive Bayes Intuition

Simple ("naive") classification method based on

Bayes rule
Relies on very simple representation of document
◦ Bag of words
 The Bag of Words Representation
it
it it 6 6
I I I 5 5
I love
I loveI love
this this
movie!
this movie!
movie!It'sIt's It's
sweet, sweet,
sweet, thethe the 4 4
fairy always love it it it
fairy
butbut
but with with
satirical
with satirical
satiricalhumor. humor.
humor. The The
The fairy
it always lovelove
always to toto to to to 3 3
dialogue itit whimsical
whimsical it itit I and and 3 3
dialogue is great
dialogue is is great
andand
great and
the thethe whimsical
and seen areare I I and
adventure scenes are fun... and
and seen areanyone seen 2
adventure
adventure scenes
scenestoare are fun...
fun... friend seen anyone
anyone seen seen2
friend dialogue
It manages
It manages to be be whimsical
whimsical friendhappy happy dialogue
dialogue yetyet yet 1 1
It manages to be whimsical happy recommend would
andand romantic
romantic while
while laughing
laughing adventure
adventure recommend
recommend would 1 1
and romantic while laughing adventure sweet of satirical
satirical would 1
whimsical
at the conventions of of
at the conventions thethe whowho sweet of movie satirical
it it whimsical 1
at thefairy
conventions
fairy tale genre. of Ithewould who itsweet
I I but
to toof
movie
romantic it timeswhimsical
times 1 1
tale genre. I would it but movie
romantic I I
fairy recommend
tale genre. itI to
recommend would
it to just about several I to
yetyet sweet times1 1
just about it
several again but romantic
the humor I sweet
humor satirical
recommendanyone. I've
it to seen
just it several
itabout several again it yet
itthe satiricalsweet1 1
anyone. I've seen several thethe again seen humor
would adventure
anyone. times,
times,I'veand and I'mI'm
seen it always
alwaysseveral happy
happy to toscenesseen the
would
scenesI Iit the manages adventure 1 1
satirical
the genre
times,
to
to and
seeseeitI'm it again
again whenever
whenever
always happy I I fun
seen
thethe the manages
times
would genre 1 1
adventure
have a friend who hasn't fun toI Iscenes and I the
times andand
manages fairy
fairy 1 1
have
to see seen a
it again friend who
whenever I hasn't and the about
abouttimeswhile while humor genre 1
it yet! fun whenever haveand humor fairy 1
haveseen it yet!who hasn't
a friend whenever Iconventions
and about have have
have 1 1
conventions
with while
seen it yet! whenever
with greathumor 1 1
have great
conventions …… have……
with great
13
 The bag of words representation
seen 2

γ( )=c
sweet 1
whimsical 1
recommend 1
happy 1
... ...
 Bayes’ Rule Applied to Documents and Classes

•For a document d and a class c

P(d | c)P(c)
P(c | d) =
P(d)
Naive Bayes Classifier (I)

MAP is “maximum a
cMAP = argmax P(c | d) posteriori” = most
c∈C likely class

P(d | c)P(c)
= argmax Bayes Rule

c∈C P(d)
= argmax P(d | c)P(c) Dropping the
denominator
c∈C
 Naive Bayes Classifier (II)
"Likelihood" "Prior"

cMAP = argmax P(d | c)P(c)

c∈C
Document d
represented as
= argmax P(x1, x2 ,…, xn | c)P(c) features
c∈C x1..xn
 Naïve Bayes Classifier (IV)

cMAP = argmax P(x1, x2 ,…, xn | c)P(c)

c∈C

O(|X|n•|C|) parameters How often does this

class occur?

Could only be estimated if a

We can just count the
very, very large number of relative frequencies in
training examples was a corpus

available.
 Multinomial Naive Bayes Independence
Assumptions
P(x1, x2 ,…, xn | c)

Bag of Words assumption: Assume position doesn’t matter

Conditional Independence: Assume the feature
probabilities P(xi|cj) are independent given the class c.

P(x1,…, xn | c) = P(x1 | c)• P(x2 | c)• P(x3 | c)•...• P(xn | c)

 Multinomial Naive Bayes Classifier

cMAP = argmax P(x1, x2 ,…, xn | c)P(c)

c∈C

cNB = argmax P(c j )∏ P(x | c)

c∈C x∈X
 Applying Multinomial Naive Bayes Classifiers
to Text Classification

positions ¬ all word positions in test document

cNB = argmax P(c j )

c j ∈C
∏ P(xi | c j )
i∈ positions
Problems with multiplying lots of probs
There's a problem with this:

cNB = argmax P(c j )

c j ∈C
∏ P(xi | c j )
i∈ positions

Multiplying lots of probabilities can result in floating-point underflow!

.0006 * .0007 * .0009 * .01 * .5 * .000008….
Idea: Use logs, because log(ab) = log(a) + log(b)
We'll sum logs of probabilities instead of multiplying probabilities!
We actually do everything in log space
Instead of this: cNB = argmax P(c j ) ∏ P(xi | c j )
c j ∈C i∈ positions
<latexit sha1_base64="o0LQfSf3I3G0xas3oLJOwQZR0GU=">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</latexit>

2 3
X
This: cNB = argmax 4log P (cj ) + log P (xi |cj )5
cj 2C
i2positions

Notes:
1) Taking log doesn't change the ranking of classes!
The class with highest probability also has highest log probability!
2) It's a linear model:
Just a max of a sum of weights: a linear function of the inputs
So naive bayes is a linear classifier
Text The Naive Bayes Classifier
Classification
and Naive
Bayes
Text
Classification Naive Bayes: Learning
and Naïve
Bayes
 Sec.13.3

Learning the Multinomial Naive Bayes Model

First attempt: maximum likelihood estimates

◦ simply use the frequencies in the data

𝑁"!
𝑃! 𝑐! =
𝑁#$#%&
count(wi , c j )
P̂(wi | c j ) =
∑ count(w, c j )
w∈V
 Parameter estimation

P̂(wi | c j ) =
count(wi , c j ) fraction of times word wi appears
∑ count(w, c j ) among all words in documents of topic cj
w∈V

Create mega-document for topic j by concatenating all

docs in this topic
◦ Use frequency of w in mega-document
 Sec.13.3

Problem with Maximum Likelihood

What if we have seen no training documents with the word fantastic

and classified in the topic positive (thumbs-up)?

count("fantastic", positive)
P̂("fantastic" positive) = = 0
∑ count(w, positive)
w∈V

Zero probabilities cannot be conditioned away, no matter the other

evidence!
cMAP = argmax c P̂(c)∏ P̂(xi | c)
i
Laplace (add-1) smoothing for Naïve Bayes

count(wi , c) +1
P̂(wi | c) =
∑ (count(w, c))+1)
w∈V

count(wi , c) +1
=
# &
%% ∑ count(w, c)(( + V
$ w∈V '
 Multinomial Naïve Bayes: Learning

• From training corpus, extract Vocabulary

Calculate P(cj) terms • Calculate P(wk | cj) terms
◦ For each cj in C do • Textj ¬ single doc containing all docsj
docsj ¬ all docs with class =cj • For each word wk in Vocabulary
nk ¬ # of occurrences of wk in Textj
| docs j |
P(c j ) ← nk + α
| total # documents| P(wk | c j ) ←
n + α | Vocabulary |
Unknown words
What about unknown words
◦ that appear in our test data
◦ but not in our training data or vocabulary?
We ignore them
◦ Remove them from the test document!
◦ Pretend they weren't there!
◦ Don't include any probability for them at all!
Why don't we build an unknown word model?
◦ It doesn't help: knowing which class has more unknown words is
not generally helpful!
Stop words
Some systems ignore stop words
◦ Stop words: very frequent words like the and a.
◦ Sort the vocabulary by word frequency in training set
◦ Call the top 10 or 50 words the stopword list.
◦ Remove all stop words from both training and test sets
◦ As if they were never there!

But removing stop words doesn't usually help

• So in practice most NB algorithms use all words and don't
use stopword lists
Text
Classification Naive Bayes: Learning
and Naive
Bayes
Text Sentiment and Binary
Classification Naive Bayes
and Naive
Bayes
gative (-), and take the following miniature training and te
Let's do a worked sentiment example!
from actual movie reviews.
Cat Documents
Training - just plain boring
- entirely predictable and lacks energy
- no surprises and very few laughs
+ very powerful
+ the most fun film of the summer
Test ? predictable with no fun
Nc
or P(c) for the two classes is computed via Eq. 4.11 as
g. We’ll Training -
use a sentimentjust plain boring
analysis domain with the Thetwo
wordclasses
with positive
doesn’t occur in the training set, so we drop
A worked sentiment example with add-1 smoothing
-
-
entirely
no
d from actual movie reviews.
predictable
surprises and
and
very
lacks
egative (-), and take the following miniature training
few
energy
andabove,
mentioned
laughs
test documents
we don’t use unknown word models for naive
lihoods from the training set for the remaining three words “pred
+ very powerful
“fun”, are as follows, from Eq. 4.14 (computing the probabilities
Cat
+ the most fun Documents
film of the summer 1. Prior from training:
Training
Test -? just with no fun of the words in the training set is left as an exercise for the reader
plain boring
predictable
- entirely predictable and lacks energy
prior P(c) for the two classes is computed via Eq. 4.11 as N : Nc
𝑃/ 𝑐 =
𝑁&!
1+1 P(-) = 3/5
- no surprises and very few laughs doc P(“predictable”|
%
𝑁)'(')*
= P(“predictable”|+) =
14 + 20 P(+) = 2/5
+ very powerful
3 2
P( ) = P(+) = 1+1 0+1
+ the most5 fun film of5the summer P(“no”| ) = P(“no”|+) =
word Test
with doesn’t? occur
predictable with no
in the training set,fun 2. (asDrop 14
so we drop it completely "with"
+ 20 9 + 20
ned above, we don’t use unknown word models for naive Bayes). The like- 0+1 1+1
rior3.
P(c) for the two classes is computed via Eq. 4.11 as Nc
: P(“fun”| ) = P(“fun”|+) =
from theLikelihoods fromthree
training set for the remaining training: Ndoc “no”, and
words “predictable”, 14 + 20 9 + 20
are as follows, from Eq. 4.14
𝑐𝑜𝑢𝑛𝑡 (computing
3 𝑤! , 𝑐 +the
1 probabilities for the
2 For the test remainder
sentence S = “predictable with no fun”, after removin
𝑝 𝑤 𝑐 = P( is)left P(+) =for the reader):
words in the training
! set
∑
= as an exercise
"∈$ 𝑐𝑜𝑢𝑛𝑡
5 𝑤, 𝑐 + |𝑉| 4. Scoring the test set:
5 the chosen class, via Eq. 4.9, is therefore computed as follows:
word with doesn’t occur 1 +the
in 1 training set, so we drop0it+completely
1 (as
P(“predictable”| ) = P(“predictable”|+) = 3 2⇥2⇥1 5
14 + 20 word models for naive9Bayes).
d above, we don’t use unknown + 20 P( like-
The )P(S| ) = ⇥ 3
= 6.1 ⇥ 10
1 1 0 1
5 34
rom the training set for the remaining three words “predictable”, “no”, and
+ +
P(“no”| ) = P(“no”|+) = 2 1⇥1⇥2
e as follows, from Eq. 4.14 14 +(computing
20 the probabilities
9 + 20 for the remainder
P(+)P(S|+) = ⇥ = 3.2 ⇥ 10 5
5 29 3
0 +as
rds in the training set is left 1 an exercise for the1reader):
+1
P(“fun”| ) = P(“fun”|+) =
14 + 20 9 + 20
Optimizing for sentiment analysis
For tasks like sentiment, word occurrence seems to
be more important than word frequency.
◦ The occurrence of the word fantastic tells us a lot
◦ The fact that it occurs 5 times may not tell us much more.
Binary multinominal naive bayes, or binary NB
◦ Clip our word counts at 1
◦ Note: this is different than Bernoulli naive bayes; see the
textbook at the end of the chapter.
 Binary Multinomial Naïve Bayes: Learning
• From training corpus, extract Vocabulary
Calculate P(cj) terms • Calculate P(wk | cj) terms
◦ For each cj in C do • Text j ¬ duplicates
Remove single docincontaining
each doc: all docsj
docsj ¬ all docs with class =cj • For
• For eacheach word
word wktype w in docj
in Vocabulary
n• ¬Retain
# of only a single instance
occurrences of w inofText
w
| docs j | k k j
P(c j ) ← nk + α
| total # documents| P(wk | c j ) ←
n + α | Vocabulary |
Binary Multinomial Naive Bayes
on a test document d
First remove all duplicate words from d
Then compute NB using the same equation:

cNB = argmax P(c j )

c j ∈C
∏ P(wi | c j )
i∈ positions

39
 without add-1 smoothing to make the differences clearer. Note that the results counts
need not be 1; the word great has a count of 2 even for Binary NB, because it appears
Binary multinominal naive Bayes
in multiple documents.

NB Binary
Counts Counts
Four original documents: + +
it was pathetic the worst part was the and 2 0 1 0
boxing scenes boxing 0 1 0 1
film 1 0 1 0
no plot twists or great scenes great 3 1 2 1
+ and satire and great plot twists it 0 1 0 1
+ great scenes great film no 0 1 0 1
or 0 1 0 1
After per-document binarization: part 0 1 0 1
it was pathetic the worst part boxing pathetic 0 1 0 1
plot 1 1 1 1
scenes satire 1 0 1 0
no plot twists or great scenes scenes 1 2 1 2
+ and satire great plot twists the 0 2 0 1
+ great scenes film twists 1 1 1 1
was 0 2 0 1
worst 0 1 0 1
 without add-1 smoothing to make the differences clearer. Note that the results counts
need not be 1; the word great has a count of 2 even for Binary NB, because it appears
Binary multinominal naive Bayes
in multiple documents.

NB Binary
Counts Counts
Four original documents: + +
it was pathetic the worst part was the and 2 0 1 0
boxing scenes boxing 0 1 0 1
film 1 0 1 0
no plot twists or great scenes great 3 1 2 1
+ and satire and great plot twists it 0 1 0 1
+ great scenes great film no 0 1 0 1
or 0 1 0 1
After per-document binarization: part 0 1 0 1
it was pathetic the worst part boxing pathetic 0 1 0 1
plot 1 1 1 1
scenes satire 1 0 1 0
no plot twists or great scenes scenes 1 2 1 2
+ and satire great plot twists the 0 2 0 1
+ great scenes film twists 1 1 1 1
was 0 2 0 1
worst 0 1 0 1
 without add-1 smoothing to make the differences clearer. Note that the results counts
need not be 1; the word great has a count of 2 even for Binary NB, because it appears
Binary multinominal naive Bayes
in multiple documents.

NB Binary
Counts Counts
Four original documents: + +
it was pathetic the worst part was the and 2 0 1 0
boxing scenes boxing 0 1 0 1
film 1 0 1 0
no plot twists or great scenes great 3 1 2 1
+ and satire and great plot twists it 0 1 0 1
+ great scenes great film no 0 1 0 1
or 0 1 0 1
After per-document binarization: part 0 1 0 1
it was pathetic the worst part boxing pathetic 0 1 0 1
plot 1 1 1 1
scenes satire 1 0 1 0
no plot twists or great scenes scenes 1 2 1 2
+ and satire great plot twists the 0 2 0 1
+ great scenes film twists 1 1 1 1
was 0 2 0 1
worst 0 1 0 1
 without add-1 smoothing to make the differences clearer. Note that the results counts
need not be 1; the word great has a count of 2 even for Binary NB, because it appears
Binary multinominal naive Bayes
in multiple documents.

NB Binary
Counts Counts
Four original documents: + +
it was pathetic the worst part was the and 2 0 1 0
boxing scenes boxing 0 1 0 1
film 1 0 1 0
no plot twists or great scenes great 3 1 2 1
+ and satire and great plot twists it 0 1 0 1
+ great scenes great film no 0 1 0 1
or 0 1 0 1
After per-document binarization: part 0 1 0 1
it was pathetic the worst part boxing pathetic 0 1 0 1
plot 1 1 1 1
scenes satire 1 0 1 0
no plot twists or great scenes scenes 1 2 1 2
+ and satire great plot twists the 0 2 0 1
+ great scenes film twists 1 1 1 1
was 0 2 0 1
Counts can still be 2! Binarization is within-doc! worst 0 1 0 1
Text Sentiment and Binary
Classification Naive Bayes
and Naive
Bayes
Text More on Sentiment
Classification Classification
and Naive
Bayes
Sentiment Classification: Dealing with Negation
I really like this movie
I really don't like this movie

Negation changes the meaning of "like" to negative.

Negation can also change negative to positive-ish
◦ Don't dismiss this film
◦ Doesn't let us get bored
Sentiment Classification: Dealing with Negation
Das, Sanjiv and Mike Chen. 2001. Yahoo! for Amazon: Extracting market sentiment from stock message boards. In
Proceedings of the Asia Pacific Finance Association Annual Conference (APFA).
Bo Pang, Lillian Lee, and Shivakumar Vaithyanathan. 2002. Thumbs up? Sentiment Classification using
Machine Learning Techniques. EMNLP-2002, 79—86.

Simple baseline method:

Add NOT_ to every word between negation and following punctuation:

didn’t like this movie , but I

didn’t NOT_like NOT_this NOT_movie but I

Sentiment Classification: Lexicons
Sometimes we don't have enough labeled training
data
In that case, we can make use of pre-built word lists
Called lexicons
There are various publically available lexicons
 MPQA Subjectivity Cues Lexicon
Theresa Wilson, Janyce Wiebe, and Paul Hoffmann (2005). Recognizing Contextual Polarity in
Phrase-Level Sentiment Analysis. Proc. of HLT-EMNLP-2005.

Riloff and Wiebe (2003). Learning extraction patterns for subjective expressions. EMNLP-2003.

Home page: https://mpqa.cs.pitt.edu/lexicons/subj_lexicon/

6885 words from 8221 lemmas, annotated for intensity (strong/weak)
◦ 2718 positive
◦ 4912 negative
+ : admirable, beautiful, confident, dazzling, ecstatic, favor, glee, great
− : awful, bad, bias, catastrophe, cheat, deny, envious, foul, harsh, hate

49
 The General Inquirer
Philip J. Stone, Dexter C Dunphy, Marshall S. Smith, Daniel M. Ogilvie. 1966. The General
Inquirer: A Computer Approach to Content Analysis. MIT Press

◦ Home page: http://www.wjh.harvard.edu/~inquirer

◦ List of Categories: http://www.wjh.harvard.edu/~inquirer/homecat.htm
◦ Spreadsheet: http://www.wjh.harvard.edu/~inquirer/inquirerbasic.xls
Categories:
◦ Positiv (1915 words) and Negativ (2291 words)
◦ Strong vs Weak, Active vs Passive, Overstated versus Understated
◦ Pleasure, Pain, Virtue, Vice, Motivation, Cognitive Orientation, etc
Free for Research Use
Using Lexicons in Sentiment Classification
Add a feature that gets a count whenever a word
from the lexicon occurs
◦ E.g., a feature called "this word occurs in the positive
lexicon" or "this word occurs in the negative lexicon"
Now all positive words (good, great, beautiful,
wonderful) or negative words count for that feature.
Using 1-2 features isn't as good as using all the words.
• But when training data is sparse or not representative of the
test set, dense lexicon features can help
Naive Bayes in Other tasks: Spam Filtering
SpamAssassin Features:
◦ Mentions millions of (dollar) ((dollar) NN,NNN,NNN.NN)
◦ From: starts with many numbers
◦ Subject is all capitals
◦ HTML has a low ratio of text to image area
◦ "One hundred percent guaranteed"
◦ Claims you can be removed from the list
Naive Bayes in Language ID
Determining what language a piece of text is written in.
Features based on character n-grams do very well
Important to train on lots of varieties of each language
(e.g., American English varieties like African-American English,
or English varieties around the world like Indian English)
Summary: Naive Bayes is Not So Naive
Very Fast, low storage requirements
Work well with very small amounts of training data
Robust to Irrelevant Features
Irrelevant Features cancel each other without affecting results

Very good in domains with many equally important features

Decision Trees suffer from fragmentation in such cases – especially if little data

Optimal if the independence assumptions hold: If assumed

independence is correct, then it is the Bayes Optimal Classifier for problem
A good dependable baseline for text classification
◦ But we will see other classifiers that give better accuracy
Slide from Chris Manning
Text More on Sentiment
Classification Classification
and Naive
Bayes
Text Classification
and Naïve Bayes

Naïve Bayes:
Relationship to
Language Modeling
Dan Jurafsky

Generative Model for Multinomial Naïve Bayes

c=China

X1=Shanghai X2=and X3=Shenzhen X4=issue X5=bonds

57
Dan Jurafsky

Naïve Bayes and Language Modeling

• Naïve bayes classifiers can use any sort of feature
• URL, email address, dictionaries, network features
• But if, as in the previous slides
• We use only word features
• we use all of the words in the text (not a subset)
• Then
• Naïve bayes has an important similarity to language
58 modeling.
Dan Jurafsky Sec.13.2.1

Each class = a unigram language model

• Assigning each word: P(word | c)
• Assigning each sentence: P(s|c)=Π P(word|c)
Class pos
0.1 I
I love this fun film
0.1 love
0.1 0.1 .05 0.01 0.1
0.01 this
0.05 fun
0.1 film P(s | pos) = 0.0000005
Dan Jurafsky Sec.13.2.1

Naïve Bayes as a Language Model

• Which class assigns the higher probability to s?

Model pos Model neg

0.1 I 0.2 I I love this fun film
0.1 love 0.001 love
0.1 0.1 0.01 0.05 0.1
0.01 this 0.01 this 0.2 0.001 0.01 0.005 0.1

0.05 fun 0.005 fun

0.1 film 0.1 film P(s|pos) > P(s|neg)
Text Classification
and Naïve Bayes

Naïve Bayes:
Relationship to
Language Modeling
Text Precision, Recall, and F1
Classification
and Naive
Bayes
 Evaluating Classifiers: How well does our
classifier work?
Let's first address binary classifiers:
• Is this email spam?
spam (+) or not spam (-)
• Is this post about Delicious Pie Company?
about Del. Pie Co (+) or not about Del. Pie Co(-)

We'll need to know

1. What did our classifier say about each email or post?
2. What should our classifier have said, i.e., the correct
answer, usually as defined by humans ("gold label")
 First step in evaluation: The confusion matrix

gold standard labels

gold positive gold negative
system system tp
positive true positive false positive precision = tp+fp
output
labels system
negative false negative true negative
tp tp+tn
recall = accuracy =
tp+fn tp+fp+tn+fn
 Accuracy on the confusion matrix
gold standard labels
gold positive gold negative
system system tp
positive true positive false positive precision = tp+fp
output
labels system
negative false negative true negative
tp tp+tn
recall = accuracy =
tp+fn tp+fp+tn+fn
 Why don't we use accuracy?
Accuracy doesn't work well when we're dealing with
uncommon or imbalanced classes
Suppose we look at 1,000,000 social media posts to find
Delicious Pie-lovers (or haters)
• 100 of them talk about our pie
• 999,900 are posts about something unrelated
Imagine the following simple classifier
Every post is "not about pie"
 Accuracy re: pie posts 100 posts are about pie; 999,900 aren't

gold standard labels

gold positive gold negative
system system tp
positive true positive false positive precision = tp+fp
output
labels system
negative false negative true negative
tp tp+tn
recall = accuracy =
tp+fn tp+fp+tn+fn
 Why don't we use accuracy?
Accuracy of our "nothing is pie" classifier
999,900 true negatives and 100 false negatives
Accuracy is 999,900/1,000,000 = 99.99%!
But useless at finding pie-lovers (or haters)!!
Which was our goal!
Accuracy doesn't work well for unbalanced classes
Most tweets are not about pie!
 Instead of accuracy we use precision and recall
gold standard labels
gold positive gold negative
system system tp
positive true positive false positive precision = tp+fp
output
labels system
negative false negative true negative
tp tp+tn
recall = accuracy =
tp+fn tp+fp+tn+fn

Precision: % of selected items that are correct

Recall: % of correct items that are selected
 ’s other
he why instead
999,900of areaccuracy we generally
tweets about somethingturn to two other
completely metrics
unrelated. showna in
Imagine
precision and recall. Precision measures the percentage of theclassifier
items that
classifier
m detected
Precision/Recall
that stupidly classified aren't
every tweet fooled
as “not aboutby the"just
pie”. This call
have 999,900(i.e.,
truethe systemand
negatives labeled
only as100positive) that arefor
false negatives in fact positive of
an accuracy (i.e.,
ive everything
according to the human negative"
gold labels).
0/1,000,000 or 99.99%! What an amazing accuracy
classifier!
Precisionlevel!
is defined
Surelyaswe should
py with this classifier? But of course this fabulous ‘no pie’ classifier would
Stupid classifier: Just
pletely useless, since it wouldn’t find say
true no:
positives
a single every tweet is "not
one of the customer comments about pie"
Precision =
100Intweets
looking• for. talk
other words, about
accuracy
true pie,
positives is +not999,900
a good
false tweets
metric
positives don't
when the goal is
over something that is rare, or at least not completely balanced in frequency,
all
s a measuresAccuracy
very• common = 999,900/1,000,000
the percentage
situation in of
theitems
world.actually = present
99.99%in the input that were
yt’s
identified
whyBut by
thethe
instead system.and
ofRecall
accuracy Recall is defined
we Precision
generally turn
foras
to this
two other metricsare
classifier shown in
terrible:
: precision and recall. Precision measures the percentage of the items that
em detected (i.e., the system labeled trueaspositives
positive) that are in fact positive (i.e.,
Recall =
itive according to the human true gold labels).
positives Precision
+ false is defined as
negatives

ision and recall will help solve true

the positives
problem with the useless “nothing is
Precision =
true positives
ssifier. This classifier, despite having+afalse positives
fabulous accuracy of 99.99%, has
e recall of 0 (since there are no true positives, and 100 false negatives, the
 needs of an application. Values of b > 1 f
A combined measure: F1
precision. When b = 1, precision and reca
t frequently used metric,
F1 is a combination and
of precision andis called Fb =1
recall.

2PR
F1 =
P+R
from a weighted harmonic mean of precisio
et of numbers is the reciprocal of the arithme
 anced;
F1anced; thisthis is the
is the most
most frequentlyused
frequently usedmetric,
metric,and
and is
is called
called FFbb=1
=1 or
orjust
justFF1 :1 :

F1 is a special case ofF the

F 1== general
2PR
2PR "F-measure" (4.16)
(4.16)
1
PP++RR
F-measure
F-measure comes
comes from
from a weightedharmonic
a weighted harmonic mean
mean of
of precision
precisionand andrecall.
recall.The
The
F-measure
harmonic mean ofis the
a set (weighted)
of numbers harmonic
is the reciprocal of the mean of
arithmetic mean of recip-
harmonic mean of a set of numbers is the reciprocal of the arithmetic mean of recip-
precision and recall
rocals:
rocals:
n
HarmonicMean(a1 , a2 , a3 , a4 , ..., an ) = 1 1
n1 1
(4.17)
HarmonicMean(a1 , a2 , a3 , a4 , ..., an ) = 1a1 + a12 + a13 + ... + an1 (4.17)
a1 + a2 + a3 + ... + an
and hence F-measure is
and hence F-measure is ✓ ◆
1 ✓ with b 2 = 1 a◆ (b 2 + 1)PR
F= 1 1 or F = (b 22 + 1)PR (4.18)
F = 1a P + (1 a) R1 or with b 2 =
1 aa F = b P+R (4.18)
a P + (1 a) R1 a b 2P + R

F1 is a special case of F-measure with β=1, α=½

 Suppose we have more than 2 classes?
Lots of text classification tasks have more than two classes.
◦ Sentiment analysis (positive, negative, neutral) , named entities (person, location, organization)

We can define precision and recall for multiple classes like this 3-way email task:

gold labels
urgent normal spam
8
urgent 8 10 1 precisionu=
8+10+1
system 60
output normal 5 60 50 precisionn=
5+60+50
200
spam 3 30 200 precisions=
3+30+200
recallu = recalln = recalls =
8 60 200
8+5+3 10+60+30 1+50+200
How to combine P/R values for different classes:
Microaveraging vs Macroaveraging

Class 1: Urgent Class 2: Normal Class 3: Spam Pooled

true true true true true true true true
urgent not normal not spam not yes no
system system system system
urgent 8 11 normal 60 55 spam 200 33 yes 268 99
system system system system
not 8 340 not 40 212 not 51 83 no 99 635
8 60 200 microaverage = 268
precision = = .42 precision = = .52 precision = = .86 = .73
8+11 60+55 200+33 precision 268+99

macroaverage = .42+.52+.86
= .60
precision 3
Text Precision, Recall, and F1
Classification
and Naive
Bayes
Text Avoiding Harms in Classification
Classification
and Naive
Bayes
 Harms of classification
Classifiers, like any NLP algorithm, can cause harms
This is true for any classifier, whether Naive Bayes or
other algorithms
Representational Harms
• Harms caused by a system that demeans a social group
• Such as by perpetuating negative stereotypes about them.
• Kiritchenko and Mohammad 2018 study
• Examined 200 sentiment analysis systems on pairs of sentences
• Identical except for names:
• common African American (Shaniqua) or European American (Stephanie).
• Like "I talked to Shaniqua yesterday" vs "I talked to Stephanie yesterday"
• Result: systems assigned lower sentiment and more negative
emotion to sentences with African American names
• Downstream harm:
• Perpetuates stereotypes about African Americans
• African Americans treated differently by NLP tools like sentiment (widely
used in marketing research, mental health studies, etc.)
Harms of Censorship
• Toxicity detection is the text classification task of detecting hate speech,
abuse, harassment, or other kinds of toxic language.
• Widely used in online content moderation
• Toxicity classifiers incorrectly flag non-toxic sentences that simply
mention minority identities (like the words "blind" or "gay")
• women (Park et al., 2018),
• disabled people (Hutchinson et al., 2020)
• gay people (Dixon et al., 2018; Oliva et al., 2021)
• Downstream harms:
• Censorship of speech by disabled people and other groups
• Speech by these groups becomes less visible online
• Writers might be nudged by these algorithms to avoid these words
making people less likely to write about themselves or these groups.
Performance Disparities
1. Text classifiers perform worse on many languages of
the world due to lack of data or labels
2. Text classifiers perform worse on varieties of even
high-resource languages like English
• Example task: language identification, a first step in NLP
pipeline ("Is this post in English or not?")
• English language detection performance worse for writers
who are African American (Blodgett and O'Connor 2017)
or from India (Jurgens et al., 2017)
Harms in text classification
• Causes:
• Issues in the data; NLP systems amplify biases in training data
• Problems in the labels
• Problems in the algorithms (like what the model is trained to
optimize)
• Prevalence: The same problems occur throughout NLP
(including large language models)
• Solutions: There are no general mitigations or solutions
• But harm mitigation is an active area of research
• And there are standard benchmarks and tools that we can use
for measuring some of the harms
Text Avoiding Harms in Classification
Classification
and Naive
Bayes