This document discusses text mining and educational discourse. It provides an overview of connecting discourse analysis and machine learning to understand language interactions. The document outlines examining theoretical connections between discourse and learning. It also discusses extracting explicit features from implicit constructs to apply machine learning for text classification and analysis. Examples are provided of part-of-speech tagging and features used for machine learning on text, including n-grams and punctuation.

1. Text Mining and
Educational Discourse
Dr. Chi-Un Lei, Dept. of Electrical and Electronic Eng.
LASI-HK 2014
(Adopted from LASI workshop 2014)
1

2. Words from the Speaker
“The key insight communicated through this
workshop is that …
If we can understand the connection between socio-
psychological processes and language by means of
the social signals encoded in them, we can
structure computational models of language
interactions more effectively.”
--- Carolyn Penstein Rosé
2

3. Outline
 Theoretical: Connection between discourse and
learning
 From rich but implicit constructs to explicit features
that capture the essence for machine learning
 Hands-on: Machine learning for text extraction
and classification
3
Automatic
Analysis
Of
Conversation
Conversational
Interventions
Positive
Learning
Outcomes

4. Educational Discourse
4
Sociolinguistics
Discourse Analysis
Language
And
Identity
Language
Use
Machine
Learning
Multi-
Level
Modeling
Applied
Statistics
Computational
Models
Of
Discourse
Analysis

5. Souffle Framework
5
Engagement Engagement
• Transactive
• Knowledge Integration
Person Person
Authority Authority
 Analysis of discussions for learning

6. Transactivity
 Building on an idea expressed earlier in a
conversation
 Using a reasoning statement
6
We don't want tmax
to be at 570 both
for the material and
[the Environment]
Well, for power and
efficiency, we want a
high tmax, but
environmentally, we
want a lower one.

7. 7

8. 8

9. System of Engagement
 Showing openness to the existence of other
perspectives
 Examples
 Nuclear is a good choice
 I consider nuclear to be a good choice
 There’s no denying that nuclear is a superior choice
 Is nuclear a good choice?
9

10. 10

11. 11

12. What is machine learning?
 Automatically or semi-automatically
 Inducing concepts (i.e., rules) from data
 Finding patterns in data (For human and computer)
 Explaining data
 Making predictions
12
Data Learning Algorithm Model
New Data
Prediction
Classification Engine

13. 13

14. Keep this picture in mind…
 Machine learning isn’t magic
 But it can be useful for identifying meaningful
patterns in your data when used properly
 Proper use requires insight into your data
 Otherwise, GIGO (Garbage In Garbage Out)
 Think like a computer!
14

15. Machine Learning for Text Mining
 Basic features: “Bag of Words”
 Represent text as a vector where each position
corresponds to a term
15
• Cows make cheese. (110010)
• Cheese make cows. (110010)
• Hamsters eat seeds. (001101)
Cheese
Cows
Eat
Hamsters
Make
Seeds

16. Basic Types of Features
 Unigram
 Single words (e.g. prefer, sandwhich)
 Bigram
 Pairs of words next to each other (e.g. eat bread)
 Simple lexical patterns
 e.g. “common denominator” versus “common multiple”
 Punctuation
 “You think the answer is 9?” vs. “You think the answer is 9.”
16

17. Part of Speech (POS) Tagging
 POS bigrams capture syntactic or stylistic
information
 e.g. “the answer which is …” vs “which is the answer”
 Pairs of POS (Part-of-Speech) tags next to each other
 DT_NN: "Determiner"_"Noun, singular or mass “
 NNP_NNP: “Proper noun, singular”_“Proper noun,
singular”
 Examples
 JJR: Adjective, comparative
17
http://www.ling.upenn.edu/courses/Fall_2003/ling001/penn_treebank_pos.html

18. Feature Space Customizations
 Machine learning algorithms look for features
that are good predictors, NOT features that
are necessarily meaningful
 Look for approximations
e.g. Don’t need to do a complete syntactic
analysis for questions
Look for question marks
Look for wh-terms that occur immediately before an
auxilliary verb --- Combined features
18

19. LightSide
 Easy UI
 Feature Extraction
 Model Building /
Machine Learning
 Error Analysis
 Data Structuring
 Free/open-source for
adoption and extension
19

20. Feature
Extraction
20

21. Machine
Learning
and
Evaluation
21

22. Recap …
“The key insight communicated through this
workshop is that …
If we can understand the connection between socio-
psychological processes and language by means of
the social signals encoded in them, we can
structure computational models of language
interactions more effectively.”
--- Carolyn Penstein Rosé
22

23. Examples of Part of Speech Tagging
1. CC Coordinating conjunction
2. CD Cardinal number
3. DT Determiner
4. EX Existential there
5. FW Foreign word
6. IN Preposition/subord
7. JJ Adjective
8. JJR Adjective, comparative
9. JJS Adjective, superlative
10.LS List item marker
11.MD Modal
12.NN Noun, singular or mass
13.NNS Noun, plural
14.NNP Proper noun, singular
15.NNPS Proper noun, plural
16.PDT Predeterminer
17.POS Possessive ending
18.PRP Personal pronoun
19.PP Possessive pronoun
20.RB Adverb
21.RBR Adverb, comparative
22.RBS Adverb, superlative23
http://www.ling.upenn.edu/courses/Fall_2003/ling001/penn_treebank_pos.html

24. Examples of Part of Speech Tagging
23.RP Particle
24.SYM Symbol
25.TO to
26.UH Interjection
27.VB Verb, base form
28.VBD Verb, past tense
29.VBG Verb,
gerund/present participle
30.VBN Verb, past participle
31.VBP Verb, non-3rd ps.
sing. present
32.VBZ Verb, 3rd ps. sing.
present
33.WDT wh-determiner
34.WP wh-pronoun
35.WP Possessive wh-
pronoun
36.WRB wh-adverb
http://www.ling.upenn.edu/courses/Fall_2003/ling001/penn_treebank_pos.html

25. Findings
 Transactivity (Berkowitz & Gibbs, 1983)
 Moderating effect on learning (Joshi & Rosé, 2007; Russell, 2005;
Kruger & Tomasello, 1986; Teasley, 1995)
 Moderating effect on knowledge sharing in working groups (Gweon et
al., 2011)
 Engagement (Martin & White, 2005)
 Correlational analysis: Strong correlation between displayed openness
of group members and articulation of reasoning (R = .72) (Dyke et al.,
in press)
 Intervention study: Causal effect on propensity to articulate ideas in
group chats (effect size .6 standard deviations) (Kumar et al., 2011)
 Mediating effect of idea contribution on learning in scientific inquiry
(Wang et al., 2011)
25