Studying the Dark Triad of Personality through Twitter Behavior - - PowerPoint PPT Presentation

Studying the Dark Triad of Personality through Twitter Behavior

Author: Daniel Preotiuc-Pietro, Jordan Carpenter, Salvatore Giorgi, Lyle Ungar Source: CIKM’16 Advisor: Jia-Ling Koh Speaker: Avon Yu Date: 2016/11/22

1

Outline

• Introduction
• Method
• Experiment
• Prediction
• Conclusion

2

Introduction

Online spaces are a medium for self-expression and social communication. Dark Triad

• Narcissism
• Machiavellianism
• Psychopathy

3

Introduction

• (Insight)Aim to directly explore the relationships

between online behaviors and the three components of the dark triad

• Build a predictive model for the dark triad traits that

uses only public Twitter information.

4

Introduction

• flow chart

5

Outline

• Introduction
• Method
• Experiment
• Prediction
• Conclusion

6

Method

Data Set

• Collected through a study on Amazon Mechanical

Turk.

• 863 Twitter users with public profiles.
• 491 Twitter users posted > 500 tokens.
• Collect recent tweets (<3200), their profile picture

and profile information.

7

Method

Dark Triad Score

Completed the ’Dirty Dozen’ questionnaire:

8

Method

Features-Text Analysis

• Input Twitter post
• Unigram

words used by at least 10% users(N=6491)

• Word Clusters

compute a word to word similarity matrix using Word2Vec. performs a dimensionality reduction using SVD. performs k-mean clustering to obtain the word clusters.

9

Method

1 2 3 4 5 0.0 1.0 2.0 3.0

(1,0) (1,1) (0,2) (2,4) (3,5)

(1,0) (3,5) (0,2) 2.23 4.24 (1,1) 1 5.38 (2,4) 4.12 1.14

1 2 3 4 5 0.0 1.0 2.0 3.0

(1,0) (1,1) (0,2) (2,4) (3,5)

10

Method

1 2 3 4 5 0.0 1.0 2.0 3.0

(1,0) (1,1) (0,2) (2,4) (3,5) (0.66,1) (2.5,4.5)

(0.66,1) (2.5,4.5)

(0,2) 1.20 3.53 (1,1) 0.34 3.80 (2,4) 3.28 0.70 (1,0) 1.05 4.74 (3,5) 4.63 0.70

1 2 3 4 5 0.0 1.0 2.0 3.0

(1,0) (1,1) (0,2) (2,4) (3,5) (0.66,1) (2.5,4.5)

11

Method

Features-Text Analysis

• LIWC

manually constructed based on psychological theory. include different parts-of-speech, topical categories and emotions.

• Sentiment & Emotions

six basic emotions: anger, disgust, fear, joy, sadness and surprise. sentiments: trust, anticipation, positive and negative sentiment.

12

Method

Features-Image Analysis

• Image Features

use state- of-the-art image recognition systems. IsDefault,Grayscale, Brightness, Contrast, Saturation, Sharpness, Blur

• Facial Features

use Face++API. Types of image: has faces, one face, multiple faces Facial presentation: ratio, glasses, posture(FacePitch,FaceRoll,FaceYaw), smile

13

Method

Platform Usage

• Profile Features
• No. tweets, tweets/day

# friends, #followers, follower–friend ratio, #listed Default background, geo-enabled Proportion and count of tweets that were retweeted or liked

14

Method

Platform Usage

• Shallow Text Features

# characters, # tokens per tweet Retweets or duplicate messages Proportion of messages which use hashtags, @-replies,@-mentions, URLs or ask for followers

15

Outline

• Introduction
• Method
• Experiment
• Prediction
• Conclusion

16

Experiment

• unigrams with the highest Pearson correlation to each
• f the dark triad traits.

17

Experiment

• Table 2:Pearson correlations between

the aggregate dark triad score and textual features extracted from tweets

• Table 3: Pearson correlations

between narcissism and textual features extracted from tweets

18

Experiment

• Table 4: Pearson correlations

between psychopathy and textual features extracted from tweets

• Table 5: Pearson correlations between

Machiavellianism and high level textual features extracted from tweets

19

Experiment

image features facial features profile features shallow text features

20

Outline

• Introduction
• Method
• Experiment
• Prediction
• Conclusion

21

Prediction

22

Outline

• Introduction
• Method
• Experiment
• Prediction
• Conclusion

23

Conclusion

• first comprehensive study using observed social

media behaviors of the dark triad.

• we managed to build a predictive model of the

three dark triad traits which achieves robust predictive performance on out-of-sample testing.

24

Thank you!

25