Author or Argumentation: Exploring the Effect of Prior Beliefs And - - PowerPoint PPT Presentation

Author or Argumentation: Exploring the Effect of Prior Beliefs And Personality Traits On Persuasiveness

Nikolay Kolyada

• Background / Motivation
• Research Questions
• Dataset
• Modelling Authors
• Modelling Interests
• Modelling Beliefs
• Modelling Personality traits
• Evaluation
• Conclusion

Outline

2

Persuasion

3

Motivation

Advertising, Marketing

4

Motivation

https://itpulse.com.ng/wp-content/uploads/2017/07/social-media.jpg

Online forums

5

Motivation

https://www.google.com/url?sa=i&source=images&cd=&ved=2ahUKEwji-rfqypbjAhVBUlAKHTNMCq0QjRx6BAgBEAU&url=https%3A%2F%2Fwww.hiveage.com%2Fblog%2Fforum-marketing-small- businesses%2F&psig=AOvVaw3AMSY07yfEy62TGvi8x1D9&ust=1562168894717718

Research Questions

○ How to represent person's interests, beliefs, and personality traits? ○ How to model the author similarity? ○ How to demonstrate the impact of the similarity of people's interests, beliefs, and personality traits on their persuasiveness, if any?

6

Motivation

Reddit

7

Constructing Dataset

SUBREDDITS 887k USERS 43m COMMENTS 3.7b

8

Constructing Dataset

Reddit

ChangeMyView

9

ChangeMyView

USERS 174k DISCUSSIONS 60k

SUBMISSIONS AVERAGE COMMENTS PER SUBMISSION 10

Constructing Dataset

Submission

Original Post Title: Cars should be equipped with both angry and apologetic horns Reply #1 … A standard and widely used method of conveying thanks and apology already exists: blinking your hazards a few times. … Reply #2 … You jump a big step here, because you don't explain why the traditional light wave/hand under mirror gesture isn't effective. I see it pretty much every time …

11

Ah, really? Didn't know that! ∆

Constructing Dataset

Submission

12

Constructing Dataset

Pairs Dataset

13

Pairs dataset

14

Constructing Dataset

Pairs dataset

15

Constructing Dataset

Pairs dataset

ORIGINAL POST ORIGINAL POST PERSUASIVE COMMENT PERSUASIVE COMMENT PERSUASIVE COMMENT NON-PERSUASIVE COMMENT OP

• subreddit_id
• author
• created
• title
• text

REPLY

• author
• created
• text
• delta

DELTA True / False 16

Constructing Dataset

Pairs dataset

TIMESPAN 01.2013 - 09.2017 8025 PERSUASIVE PAIRS 8025 NON-PERSUASIVE PAIRS

17

Constructing Dataset

Argumentation features

18

• Linguistic features

BOW, articles, pronouns, links, positive/negative words, question marks, quotations

• Interplay features

ORIGINAL POST ORIGINAL POST PERSUASIVE COMMENT PERSUASIVE COMMENT PERSUASIVE COMMENT NON-PERSUASIVE COMMENT

Constructing Dataset

Argumentation features

19

Constructing Dataset

• Linguistic features
• Interplay features

Modelling Author Interests

20

21

Modelling Author Interests

Modelling Interests

22

Modelling Author Interests

Modelling Interests

Discussion Hobbies News and Politics Entertainment Science

12 91 32 ...

POST FREQUENCIES TF-IDF PCA 23

Modelling Author Interests

Modelling Interests

Discussion Hobbies News and Politics Entertainment Science

Modelling Author Beliefs

24

25

Modelling Author Beliefs

Modelling Author Beliefs

Context Sentiment

26

Modelling Author Beliefs category tagret post text context sentiment News And Politics –> Republican Hillary Clinton ... Besides, Hillary Clinton also has some weaknesses. ...

• 0.362

27

Modelling Author Beliefs

Modelling Author Beliefs

28

Modelling Author Beliefs Discussion Hobbies News and Politics Entertainment Science

Modelling Author Beliefs

29

Modelling Author Beliefs

• 0.6
• 0.2

0.6 0.2 ...

MEDIAN SENTIMENT SCORE WITHIN TOPICS

Donald Trump Hillary Clinton BREXIT Second Amendment

Discussion Hobbies News and Politics Entertainment Science

Modelling Author Beliefs

Modelling Personality Traits

30

Big Five

31

Modelling Personality Traits

Openness

Conscientiousness

Neuroticism Agreeableness Extraversion PERSONALITY

OCEAN, CANOE

32

Modelling Personality Traits

Modelling Personality Traits

33 awful, though, lazy, worse, depressing, irony, road, terrible, stressful, horrible neuroticism

• ther,

drinks, restaurant, dancing, cats, grandfather, countless, shots, computer, glorious extraversion folk, humans, poet, art, universe, narrative, giveaway, century, sexual, films

• penness

agreeableness wonderful, together, visiting, morning, spring, walked, beautiful, staying, felt, cost completed, adventure, stupid, boring, adventures, desperate, enjoying, saying, utter, extreme conscientiousness BIG FIVE

Modelling Personality Traits

Modelling Personality Traits

34

12 91 32 ...

BIG FIVE

12 91 32 ...

TRAITS PERCENTILES

12 91 32 ...

WORD CATEGORIES

Modelling Personality Traits

awful, though, lazy, worse, depressing, irony, road, terrible, stressful, horrible neuroticism

• ther,

drinks, restaurant, dancing, cats, grandfather, countless, shots, computer, glorious extraversion folk, humans, poet, art, universe, narrative, giveaway, century, sexual, films

• penness

agreeablenes s wonderful, together, visiting, morning, spring, walked, beautiful, staying, felt, cost completed, adventure, stupid, boring, adventures, desperate, enjoying, saying, utter, extreme conscientiousnes s BIG FIVE

Modelling Personality Traits

Author Similarity

35

Author Similarity

12 91 32 ...

POLITICS

12 91 32 ...

RELIGION

12 91 32 ...

HOBBIES

12 91 32 ...

POLITICS

12 91 32 ...

RELIGION

12 91 32 ...

HOBBIES 36

Author Similarity

ORIGINAL POST PERSUASIVE COMMENT PERSUASIVE COMMENT PERSUASIVE COMMENT

Author Similarity

POLITICS POLITICS POLITICS 37

Author Similarity

• 0.6
• 0.2

0.2 ... 0.6 12 91 32 ... 12 91 32 ...

Author Similarity

TOTAL OVER 7 000 FEATURES cosine( "haudpe", "TheWildThing" )

38

Author Similarity

POLITICS POLITICS POLITICS

• 0.6
• 0.2

0.2 ... 0.6 12 91 32 ... 12 91 32 ...

Feature groups:

• interests
• beliefs
• personality traits
• argumentation features

Features overview

39

40

○ How to demonstrate the impact of the similarity of people's interests, beliefs, and personality traits on their persuasiveness, if any?

Predicting the Persuasive Argument Opinion Malleability Prediction

41

Predicting the Persuasive Argument

12 91 32 ...

PERSUASIVE OR NON- PERSUASIVE

42 Interests

12 91 32 ...

Beliefs

12 91 32 ...

Traits

12 91 32 ...

Argumentation

12 91 32 ...

Interests

12 91 32 ...

Beliefs

12 91 32 ...

Traits

12 91 32 ...

Argumentation ORIGINAL POST PERSUASIVE COMMENT PERSUASIVE COMMENT PERSUASIVE COMMENT ORIGINAL POST NON- PERSUASIVE COMMENT

Evaluation

PERSUASIVE OR NON- PERSUASIVE

Predicting the Persuasive Argument

43

Evaluation

How easy is to persuade OP?

44

Opinion Malleability

45

Evaluation

Opinion Malleability

RESISTANT OR MALLEABLE personality traits BOW

ORIGINAL POST

Evaluation

46

12 91 32 ... 12 91 32 ...

Opinion Malleability

POSTS 8 121 MALLEABLE 3 186 UNIQUE AUTHORS 4 641

47

Evaluation

Opinion Malleability

48

Evaluation

Opinion Malleability

49

Evaluation

Future work

• structure of arguments, quality of arguments
• correlation between author traits and argumentation used
• model the experience: deltas earned
• different discussion platforms

50

Summary

51

• Research questions

○ How to represent person's interests, beliefs, and personality traits? ○ How to model the author similarity? ○ How to demonstrate the impact of the similarity of people's interests, beliefs, and personality traits on their persuasiveness, if any?

• Pairs dataset
• Author features: Interests, Beliefs, Personality Traits
• Experiments: Prediction of persuasive argument, Opinion malleability
• Explore the impact of author similarity on persuasion