Michael Young www'socialscienceautomation'com Turning Text Into Data

1
Michael Youngwww.socialscienceautomation.com
Turning Text Into Data
2
The lure

• The formal approach to discourse analysis
  undertaken by Social Science Automation would
  undoubtedly be of interest to many students and
  faculty in IR, both from a practical standpoint
  in that it provides a useful tool for conducting
  research, and from a methodological perspective,
  representing an innovative mix of content and
  discourse analysis that seeks to capture formally
  the meanings of different texts.

3
Agenda

• Why Social Science Automation exists
• Brief overview of WorldView
• Slightly longer overview of Profiler Plus and
  what it does.
• Does it work?
• Some current projects
• Demonstrations
• Please ask questions at any time.

4
Why Social Science Automation Exists

• Most of the available data in the social sciences
  is text.
• For most of the history of the social sciences we
  have dealt with text extraordinarily poorly.
• In 1997, I was a failing academic ...
• Many current and former faculty here supported my
  efforts in many ways

5

• These circumstances led to the development of
  WorldView and later of Profiler Plus (because
  WorldView turned out to be the easy part).

6
WorldView

• WorldView is a program for representing,
  exploring, and comparing semantic network
  representations of text.

7
For the United States to be strong abroad, it
must be strong at home.
8
Profiler Plus

• Profiler Plus is a general purpose text coding
  program.
• That is, a rule-based system for manipulating
  tokens (words and punctuation), whose rules are
  collected in sets and applied in sequence.

9
Profiler Plus
Data
Coding Scheme
10
JARGON!

• Token word or punctuation (hi ! ,).
• Slot a place in a token to store information.
• Literal the sequence of actual characters found
  in the text.
• Part of speech grammatical function of a token,
  noun.
• Lemma base form of a word.
• lemmings?lemming

11
Definition of A Coding Scheme

• A coding scheme consists of a set of rules for
• Adding, deleting and moving tokens
• Changing the values in a token's slots
• Outputting results
• All based on the tokens other slot values and
  relationships to other tokens
• The simplest case is matching a literal.
• ? find all occurrences of anger and output
  ANGER for each occurrence.

12
Sequence of Operations

• Tokenizing
• Building sentence data structures
• Assigning a part of speech and lemma to each
  token
• Manipulating the tokens
• Creating output

13
Tokenizing

• Breaks the text into sentences which limits the
  scope of token matching to a single sentence.
• Separates punctuation from words so that we do
  not have to concern ourselves with trailing
  periods, commas, semicolons and etc .

14

• Sentence data structure
• Rome was destroyed by the Huns

15

• Part of Speech
• Noun, verb, preposition, adjective, adverb
• Lemma
• Base form of token

16

• Rome was destroyed by the Huns

17

• Manipulating tokens
• Using rules with operators that create, move, and
  delete tokens as well as comparing and changing
  values of the slots in the tokens.

18
Reduction Table
19

• Rome has been destroyed by the Huns.
• ?
• Huns destroyed Rome.

20
Does it work?
21
To quote a Harvard Professor...
Dont take my word for it...

• Our results surprised us. The computer program we
  evaluated VRA Reader was able to extract
  information on a level equal to trained
  Harvard undergraduatesand this was for a short
  term application. Computer programs do not get
  tired, bored, and distracted , perhaps for the
  first time, automated programs are ready for
  primetime.
• Gary King Will Lowe, International Organization
  57, 2003

22
(No Transcript)
23
(No Transcript)
24

25
Select Projects

• ICEWS-CWEST (DARPA)
• Near real-time event data extraction from the
  full text of news feeds
• Coding capabilities in Arabic and Russian
• Perception Metrics
• Company branding
• Nation branding
• Political Perception
• Candidate image in media
• Achievement Metrics
• Who are the trouble makers?
• Refinement of text mapping and WorldView

26
Demonstrations

• XEditor
• Profiler Plus
• WorldView