Interplay Between Dominant Personality Traits with Tie Formation in

Interplay Between
Dominant Personality Traits with
Tie Formation in a
Corporate Communication Network
Craig Evans (Presenter), Rezvaneh Rezapour, Ming Jiang, Jana Diesner
Sunbelt XXXVI (5-10 April 2016), Newport Beach, California
Motivation – Methodological:
Combining Natural Language Processing
(NLP) with Social Network Analysis helps to:
Consider content and structure of social relationships
(Lazer et al. 2009)
Understand mutual impact of language use and
network formation (Milroy 1987, Roth & Cointet 2010)
Enhance social network data with node (e.g.
subjectivity) and edge properties (e.g. type, valence)
that are difficult or expensive to obtain otherwise
(Diesner & Evans 2015)
illinois informatics institute
GSLIS - the iSchool at Illinois
Motivation – Theoretical:
Testing homophily theory (McPherson, SmithLovin & Cook 2001) requires node attributes
Reliable measurement: socio-demographic
properties, health-related information (Christakis &
Fowler 2007)
More costly to measure: information via interviews,
questionnaires (self-reported data), e.g. personal
values
Personality traits of members of business
organizations impact organizations (corruption)
and vice versa (Aven 2010, Pinto et al 2008)
How can this be measured?
illinois informatics institute
GSLIS - the iSchool at Illinois
Research Question
Is there a tendency for people with similar
personal values to form links and clusters
in communication networks (homophily)?
illinois informatics institute
GSLIS - the iSchool at Illinois
Background: Moral Judgment
Method
User studies
(Interviews,
questionnaire, phone
survey)
User study + neurophysiological
measures (Decety et
al. 2012)
Key Findings
Moral judgment correlates with:
Age (Rest, 1974)
Education (Rest, Narvaez, Thoma, & Bebeau, 1999)
Gender (males slightly more justice oriented, females more care oriented (Jaffee &
Hyde, 2000)
Culture (people from Eastern cultures are stronger concerned with ingroup and
purity than Westerners (Graham et al., 2011))
Political orientation (liberals more concerned with fairness/ unfairness and honesty/
dishonesty, conservatives with loyalty/disloyalty and sancity/degradation (Hofmann,
Wisneski, Brandt, & Skitka, 2014)
Morality is contagious (Hofmann et al., 2014)
-
NLP (Sagi & Dehghani, 2014)
-
People of all ages able to identify wrongness, intentional harm perceived more
wrong than accidental harm (visual analogue scale, VAS)
Ability to differentiate between accidental and intentional harm and harming an
object versus a person increase with age
News coverage addresses all dimensions of moral (as per MFT) for controversial
topics (MFQ)
Moral judgment correlates with political viewpoints: Democrats identify stronger
with fairness, Republicans with purity
illinois informatics institute
GSLIS - the iSchool at Illinois
Background: Measuring Personal
Values
Existing studies of personal moral:
Controlled lab experiments
Limitations
Limited Scalability
Artificial Environment with no consequence of actions
Unreliable self reported data
illinois informatics institute
GSLIS - the iSchool at Illinois
Archival Data: Emails
Explicit social network can be constructed
from “envelope” that details who it was
sent from and to
Date and time stamped (enables longitudinal
analysis)
Social network can be enriched with
information from substance/bodies of
emails
Relational properties
Link type, Link valence (via sentiment analysis)
Node properties (e.g. personality traits)
illinois informatics institute
GSLIS - the iSchool at Illinois
Method
Enron
eMails
Text
Parsing,
Extraction
Tokenized
words,
POS, stem
and lemma
Lexical
Resource
(MFD)
Term
Expansion
Synonym/
Antonym
Expansion
Queries
WordNet
Dictionary
JWNI
Analysis
illinois informatics institute
Database
GSLIS - the iSchool at Illinois
Method: Moral Foundations Theory
Dimensions
Virtue
Vice
Care
Harm
Fairness
Cheating
Loyalty
Betrayal
Explanation
Protecting versus hurting others
Cooperation / trust versus
cheating in interactions with
objects and people
Ingroup commitment (to
coalitions, teams, brands)
versus leaving group
Authority
Subversion
Playing within the rules of
hierarchy versus challenging
hierarchies
Sanctity
Degradation
Behavioral immune system
versus spontaneous reaction
illinois informatics institute
GSLIS - the iSchool at Illinois
Method: Moral Foundations
Theory: Operationalization
Based on Moral Foundations Theory (Graham,
Haidt et. al 2012)
Previously operationalized as Moral
Foundations Dictionary
Theoretical/domain/subject matter expertise
Concise
Available as LIWC dictionary
1 www.moralfoundations.org
illinois informatics institute
GSLIS - the iSchool at Illinois
Method: Moral Foundations
Operationalization: Expansion
Dictionary Refinement Operation
Original MFD Size
Words (all Categories)
Words (Unique)
Wildcard Expansion
Pruning
WordNet Expansion
Additional Synonyms
Additional Antonyms
Total Added
After Removing Redundant Words
After Removing Stopwords
Word
Count
359
324
2,368
1,408
4,703
981
5,684
4,524
4,339
Original baseline Moral Foundations Dict
Using WordNet dictionary files,
expand words AND obtain POS
Manual pruning of irrelevant Word/POS
Expand dictionary with WordNet via
JWNI
Manual pruning of irrelevant Word/POS
Final Dictionary
Virtues
Vices
General
Total
illinois informatics institute
1,879
1,878
582
4,339
GSLIS - the iSchool at Illinois
Communication Matrix
Sender/Receiver – Period 1 !
Receiver
Authority Subversion Care Harm
Authority
Cheating Loyalty Betrayal Sanctity Degradation UNKNOWN
Total
2666
1
123
78
1
3
465
62
4
0
4429
7832
1
0
0
0
0
0
0
0
0
0
2
3
Care
131
0
7
2
0
0
34
4
1
0
586
765
Harm
89
0
7
1
1
0
22
3
0
0
191
314
Fair
1
0
0
0
0
0
1
0
0
0
0
2
Cheating
2
0
0
0
0
0
0
0
0
0
0
2
Loyalty
586
0
31
19
0
0
138
11
1
0
1503
2289
Betrayal
65
0
0
2
0
1
10
1
0
0
82
161
Sanctity
5
0
0
0
0
0
3
0
0
0
3
11
Degradation
0
0
0
0
0
0
0
0
0
0
0
0
NEUTRAL
695
0
42
37
1
0
223
28
3
0
4924
5953
Total
4241
1
210
139
3
4
896
109
9
0
11720
Subversion
Sender
Fair
illinois informatics institute
GSLIS - the iSchool at Illinois
Communication Matrix
Sender/Receiver – Period 2 "
Receiver
Authority Subversion Care Harm
Authority
Cheating Loyalty Betrayal Sanctity Degradation UNKNOWN
Total
1720
5
114
48
0
0
307
39
12
0
3700
5945
1
0
0
0
0
0
0
0
0
0
0
1
Care
108
0
10
6
0
0
29
5
0
0
838
996
Harm
43
0
2
0
0
0
5
1
2
0
94
147
Fair
0
0
0
0
0
0
0
0
0
0
0
0
Cheating
1
0
0
0
0
0
0
0
0
0
1
2
Loyalty
182
0
15
7
0
0
52
2
0
0
626
884
Betrayal
41
0
4
2
0
0
11
1
0
0
165
224
Sanctity
10
0
0
0
0
0
0
0
2
0
2
14
Degradation
0
0
0
0
0
0
0
0
0
0
0
0
NEUTRAL
513
0
98
33
0
0
347
16
12
0
6097
7116
Total
2619
5
243
96
0
0
751
64
28
0
11523
Subversion
Sender
Fair
illinois informatics institute
GSLIS - the iSchool at Illinois
Communication Matrix
Sender/Receiver – Period 3 #
Receiver
Authority Subversion Care Harm
Authority
Cheating Loyalty Betrayal Sanctity Degradation UNKNOWN
Total
2957
7
136
74
1
1
449
53
0
0
5239
8917
17
0
0
0
0
0
1
0
0
0
5
23
Care
155
0
14
7
0
1
35
2
0
0
945
1159
Harm
177
0
7
5
0
0
26
3
0
0
344
562
Fair
1
0
0
0
0
0
0
0
0
0
0
1
Cheating
4
0
0
0
0
0
0
0
0
0
0
4
Loyalty
404
0
40
9
0
2
85
7
0
0
1281
1828
Betrayal
84
0
12
1
0
2
15
3
0
0
242
359
Sanctity
0
0
0
0
0
0
0
0
0
0
0
0
Degradation
0
0
0
0
0
0
0
0
0
0
0
0
NEUTRAL
605
5
84
39
0
1
182
22
0
0
5072
6010
Total
4404
12
293
135
1
7
793
90
0
0
13128
Subversion
Sender
Fair
illinois informatics institute
GSLIS - the iSchool at Illinois
Communication Matrix
Sender/Receiver – Period 4 $
Receiver
Authority Subversion Care Harm
Sender
Authority
Fair
Cheating Loyalty Betrayal Sanctity Degradation UNKNOWN
Total
2244
0
74
116
6
0
555
50
4
0
5715
8764
Subversion
0
0
0
0
0
0
0
0
0
0
0
0
Care
84
0
10
1
0
0
37
3
0
0
317
452
Harm
126
0
3
4
0
0
33
3
0
0
218
387
Fair
4
0
0
0
0
0
0
0
0
0
1
5
Cheating
0
0
0
0
0
0
0
0
0
0
0
0
Loyalty
411
0
12
20
2
0
79
9
1
0
763
1297
Betrayal
107
0
2
14
2
0
19
4
1
0
235
384
Sanctity
1
0
0
0
0
0
0
0
0
0
2
3
Degradation
0
0
0
0
0
0
0
0
0
0
0
0
NEUTRAL
548
0
54
59
1
0
249
41
1
0
5345
6298
Total
3525
0
155
214
11
0
972
110
7
0
12596
illinois informatics institute
GSLIS - the iSchool at Illinois
Communication Matrix
Sender/Receiver – Period 5 %
Receiver
Authority Subversion Care Harm
Authority
Cheating Loyalty Betrayal Sanctity Degradation UNKNOWN
Total
4701
21
191
203
36
24
1688
127
22
0
7392
14405
9
0
2
2
0
0
6
0
0
0
11
30
Care
178
3
6
6
2
1
69
5
0
0
379
649
Harm
231
1
10
9
0
0
102
6
0
0
508
867
Fair
15
0
0
0
0
0
4
2
0
0
6
27
Cheating
10
0
1
0
0
0
2
0
0
0
8
21
Loyalty
1449
9
53
68
7
5
588
37
7
0
2824
5047
Betrayal
235
0
7
8
4
1
71
16
0
0
356
698
Sanctity
10
0
0
2
0
0
8
0
0
0
25
45
Degradation
0
0
0
0
0
0
0
0
0
0
0
0
NEUTRAL
1305
8
48
88
1
1
792
27
6
0
4165
6441
Total
8143
42
318
386
50
32
3330
220
35
0
15674
Subversion
Sender
Fair
illinois informatics institute
GSLIS - the iSchool at Illinois
Distribution of Dominant Traits
illinois informatics institute
GSLIS - the iSchool at Illinois
Do Birds of a Feather flock?
Homophily in Period 1 !
Betrayal
Sanctity
Degradation
Receiver
LoyaltyTrait
Loyalty
Subversion
Care
Harm
Fair
Cheating
Authority
Harm
Authority
Subversion
Betrayal
Sanctity
Degradation
UNKNOWN
Personality Traits − Sender to Receiver
Fairness
Receiver
Care1: 1 December
Period
Cheating 2000 − 28 February 2001
Unknown
Pearson
residuals:
17.0
Authority
7.5
Subversion
Subversion
CareCare
Harm
Harm
Fair
Cheating
Fairness
Cheating
Loyalty
Loyalty
Sender Trait
Sender
Authority
Sanctity
Degradation
Betrayal
Betrayal
Sanctity
Degradation
1.0
0.0
−1.0
−7.5
Neutral
NEUTRAL
−20.0
p−value =
< 2.22e−16
illinois informatics institute
GSLIS - the iSchool at Illinois
Do Birds of a Feather flock?
Homophily in Period 2 "
Personality Traits − Sender to Receiver
Fairness
Receiver
Care
Period 2: Cheating
1 March 2001
− 30 April 2001
Betrayal
Sanctity
Degradation
UNKNOWN
Betrayal
Sanctity
Degradation
LoyaltyTrait
Receiver
Loyalty
Authority
Subversion
Care
Harm
Fair
Cheating
Harm
Authority
Subversion
Unknown
Pearson
residuals:
22.0
Authority
Subversion
Subversion
Harm
Fairness
Cheating
Sanctity
Degradation
Care Care
Sender Trait
Sender
Authority
7.5
Harm
Fair
Cheating
LoyaltyLoyalty
Betrayal
Betrayal
Sanctity
1.0
−1.0
Degradation
−7.5
NEUTRAL
Neutral
−20.0
p−value =
< 2.22e−16
illinois informatics institute
GSLIS - the iSchool at Illinois
Do Birds of a Feather flock?
Homophily in Period 3 #
LoyaltyTrait
Receiver
Betrayal
Sanctity
Degradation
Betrayal
Sanctity
Degradation
Loyalty
Authority
Subversion
Care
Harm
Fair
Cheating
Harm
Authority
Subversion
UNKNOWN
Personality Traits − Sender to Receiver
Fairness
Receiver
Care
Period 3: 1
May 2001
− 15 August 2001
Cheating
Unknown
Pearson
residuals:
19.0
7.5
Subversion
Subversion
CareCare
Harm
Harm
Fair
Cheating
Fairness
Cheating
Loyalty
Loyalty
Betrayal
Betrayal
Sanctity
Degradation
Sanctity
Degradation
Sender Trait
Sender
Authority
Authority
1.0
−1.0
−7.5
Neutral
NEUTRAL
−21.0
p−value =
< 2.22e−16
illinois informatics institute
GSLIS - the iSchool at Illinois
Do Birds of a Feather flock?
Homophily in Period 4 $
Loyalty
Receiver Trait
Betrayal
Sanctity
Degradation
Betrayal
Sanctity
Degradation
Loyalty
Authority
Subversion
Care
Harm
Fair
Cheating
Authority
Subversion
UNKNOWN
Personality Traits − Sender to Receiver
Fairness
Receiver
Care
Period
4:
16
August
2001
− 15 October 2001
Cheating
Harm
Unknown
Pearson
residuals:
12.0
Subversion
Subversion
CareCare
Harm
Harm
Fair
Cheating
Fairness
Loyalty
Cheating Loyalty
Sender Trait
Sender
Authority
Authority
Sanctity
Degradation
7.5
1.0
0.0
−1.0
−7.5
Betrayal
Betrayal
Sanctity
Degradation
Neutral
NEUTRAL
−20.0
p−value =
< 2.22e−16
illinois informatics institute
GSLIS - the iSchool at Illinois
Do Birds of a Feather flock?
Homophily in Period 5 %
Personality Traits − Sender to Receiver
Fairness
Receiver
Care5: 16 October
Period
Cheating 2001 − 31 December 2001
Betrayal
Sanctity
Degradation
Receiver
Trait
Loyalty
Betrayal
Sanctity
Unknown
Degradation
UNKNOWN
Loyalty
Authority
Subversion
Care
Harm
Fair
Cheating
Harm
Authority
Subversion
Pearson
residuals:
9.80
Sender Trait
Sender
Authority
Authority
Subversion
Subversion
Care
CareHarm
Harm
Fair
Cheating
Fairness
Cheating
7.50
0.75
0.00
−0.75
LoyaltyLoyalty
−7.50
Sanctity
Degradation
Betrayal
Betrayal
Sanctity
Degradation
NEUTRAL
Neutral
illinois informatics institute
−13.00
p−value =
< 2.22e−16
GSLIS - the iSchool at Illinois
Conclusion
Trait to trait correlation exists …
but it isnt necessarily homophily
Homophily with respect to personal values
in a corporate communication networks
exists
Authority-Authority is strong
Loyalty-Loyalty, Harm-Harm is moderate
Other trait to trait communications
correlate to varying degrees
Authority-Loyalty and Loyalty-Authority is moderate
illinois informatics institute
GSLIS - the iSchool at Illinois
Limitations and Next Steps
Limitation
The primary personality trait expressed in an email
may not accurately reflect the actual personality of
the individual
Next Steps –
Evaluation against a comparable dataset
(LDC Avocado)
Integration with ongoing tie valence research
Integration with ongoing triad and clique research
Evaluate EI Index as a comparator to statistical
correlation
illinois informatics institute
GSLIS - the iSchool at Illinois
Questions/Comments and Contact
Questions:
Contact:
Craig Evans
Jana Diesner
illinois informatics institute
- [email protected]
- [email protected]
GSLIS - the iSchool at Illinois