Scribd
Upload
485 views

Computer Assisted Text Analysis

The document describes various computer-assisted text analysis techniques that can be used for exploratory or hypothesis-driven analysis of textual data. It outlines unsupervised methods like topic modeling and lexical selection that can be used to explore themes or styles within text without predefined categories. It also discusses supervised machine learning methods and dictionaries that can be used to test hypotheses about themes or styles when categories are predefined. Finally, it provides examples of how natural language processing techniques like part-of-speech tagging, named entity recognition, and concordances can help test specific linguistic hypotheses.

Uploaded by

Terry College of Business
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
485 views

Computer Assisted Text Analysis

The document describes various computer-assisted text analysis techniques that can be used for exploratory or hypothesis-driven analysis of textual data. It outlines unsupervised methods like topic modeling and lexical selection that can be used to explore themes or styles within text without predefined categories. It also discusses supervised machine learning methods and dictionaries that can be used to test hypotheses about themes or styles when categories are predefined. Finally, it provides examples of how natural language processing techniques like part-of-speech tagging, named entity recognition, and concordances can help test specific linguistic hypotheses.

Uploaded by

Terry College of Business
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 37

Computer-Assisted Text

Analysis:
An Overview and Guide
Laura K. Nelson
Kellogg School of Management
Content Analysis PWD
Academy of Management Annual
Conference
August 7, 2015
Vancouver, BC

Goals
Describe the wide range of

computer-assisted text analysis


techniques available
Hint: more than dictionaries

Provide some guidance about

how to choose your methods


Give empirical examples of
these methods in practice

Why Use Computer-Assisted


Techniques?
Speed
Humans are slow
Text is becoming large
Reliability / Reproducibility
Validity (this is controversial)
Expanded memory
Unburdened by bias

Does not remove the need for


interpretation!

Overview:
Types of Automated Text Analysis
Unsupervised exploration (hypothesis

forming/inductive)
Topic modeling
Lexical selection

Human-Guided Categorical Analysis (traditional

content analysis deductive hypothesis testing)


Supervised machine learning
Dictionaries

Natural Language Processing (guided

inductive/hypothesis testing)
Part-of-Speech Tagging
Named Entity Recognition
Concordances
Sentiment analysis

Question 1:
Do you want to inductively explore the
text?

Unsupervised Exploration:
The Goal

Informative
Groups of
Words

Set-Up: Document-Term
Matrix*
ambit povert people full
i
Docume 4
2
0
0
nt1
Docume 1
3
7
0
nt2
Docume 2
0
0
1
nt3
*Cells
can 9be word1 frequencies
or 2
Docume
4
weighted
word scores
nt4

Question 2: Themes or Style?


If themes:
Question 3: Multiple
Categories?

Single Category per Text:


Clustering

Multiple Categories: Topic


Modeling

But Which Algorithm?


Is the order of the documents

important?
Yes? Structural Topic Modeling (STM)
Are the topics correlated?
Yes? Correlated Topic Modeling (CTM)
Order is relatively arbitrary, topics
may not be related?
Latent Dirichlet Allocation (LDA)

Question 2: Themes or Style?

If style

Style: Lexical Selection


Goal: find words that are

distinctive to different groups


of text
One solution: Difference of
Proportions

Difference of Proportions
Chicago
chicago
children
center
union
school
abort
nixon
day
vietnam
people
city
hospital
cwlu

Abstract

DoP
5.31
4.59
4.34
3.61
3.48
3.19
2.93
2.86
2.57
2.50
2.44
2.38
2.37

New York City


DoP
movement
12.54
women
11.34
feminist
8.91
radical
8.56
liberation
7.69
political
5.81
history
5.68
feminine
3.85
male
3.52
left
2.96
revolution
2.58
consciousnessraising
2
oppress
2.41

Concret
e

Question 1:
Do you want to test a
hypothesis?
If yes:
Question 2: Themes or
Styles?
If themes

Supervised Machine Learning

Which Algorithm?
You want individual

documents coded:
Document Classification
(e.g. SVM, Naive Bayes)
You want proportion of
documents in each category:
ReadMe (R package)

Dictionary Methods

Standardized

Dictionaries
LIWC (can be used for
sentiment analysis)
Custom Dictionary

Question 2: Themes or Styles?

If styles

Natural Language Processing

Takes into account features

of words, relationships
between words,
grammatical structures, etc.

Examples: Use NLP to test


hypotheses
Hypothesis: Author A is more

descriptive than Author B.


Test: Part-of-Speech tagger,
extract adjectives, count and
compare.
Hypothesis: Organizations in New
York City are more internationally
focused than organizations in
Silicon Valley.
Test: Named Entity Recognition,

Example: Use NLP to test


hypotheses
Hypothesis: the word disruptive

is used in a positive way, and has a


different meaning, for Silicon Valley
organizations compared to Wall
Street organizations.
Test: Concordances

NLP: Concordances
ong the former , one was of a most
ON OF THE PSALMS . " Touching that
ll over with a heathenish array of
d as you gazed , and wondered what
that has survived the flood ; most
they might scout at Moby Dick as a
th of Radney .'" CHAPTER 55 Of the
ing Scenes . In connexion with the
ere to enter upon those still more
ght have been rummaged out of this

monstrous
monstrous
monstrous
monstrous
monstrous
monstrous
monstrous
monstrous
monstrous
monstrous

size . ... This came towards us ,


bulk of the whale or ork we have r
clubs and spears . Some were thick
cannibal and savage could ever hav
and most mountainous ! That Himmal
fable , or still worse and more de
Pictures of Whales . I shall ere l
pictures of whales , I am strongly
stories of them which are to be fo
cabinet there is no telling . But

very heartily so exceedingly remarkably as vast a great

amazingly extremely good sweet


Mostly positive

mean part maddens doleful gamesome subtly

uncommon careful untoward exasperate loving passing


mouldy christian few true mystifying imperial modifies
contemptible
Mostly negative

Example: NLP and WordNet


Hypothesis: Womens movement

organizations in New York City approach


politics more abstractly compared to those
in Chicago, who have a more concrete
approach to politics.

Tactics and Issues Over Time


Structural Topic Models (structured on

year)
Used R package stm (Roberts, Stewart, and Tingly)
Further grouped the 40 topics into 7
topic categories
Python NLTK, extracted verbs/verb
phrases
Hand identified tactics, created
dictionaries of tactical categories

Percent Total Words

Percent Total Words in 2000

Percent Total Words in 2012

Tactics by Year

Tactics by Year

Conclusion:
Research design is key! Good data

is critical!
Match your method to your
question and data. Be purposeful,
not trendy
Use multiple methods, including
qualitative, to verify the analysis
Learn a programming language
Off-the-shelf tools box you in (see
point 2).

Happy text
analyzing!

Laura K. Nelson
laura.nelson@kellogg.northw
estern.edu

Tactical Categories*
Direct Environmental Protection: build, improve, protect, recycle
Non-Disruptive Protest: chant, demonstrate, organize, petition, protest
Disruptive Protest: blockade, chain, prevent, damage, sabotage
Political: campaign, donate, elect, endorse, regulate
Juridical: audit, enforce, inspect, represent, testify
Verbal Statements: advocate, comment, criticize, explain, refute
Business: boycott, buy, invest, purchase, sponsor
Education/Raising Awareness: editorial, outreach, publish, report, tweet
Organization/Movement Building: fund-raise, initiate, launch, participate
Negotiations: deal, discuss, engage, listen, persuade
*Categories are not mutually exclusive

You might also like