๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ

ML/Papers

[WWW 2013] Information Sharing on Twitter During the 2011 Catastrophic Earthquake ๋ฆฌ๋ทฐ

์ด๋ฒˆ์— ์†Œ๊ฐœํ•  ๋…ผ๋ฌธ์€ "Information Sharing on Twitter During the 2011 Catastrophic Earthquake " ์ž…๋‹ˆ๋‹ค.

 

0. ABSTRACT

1. INTRODUCTION

2. RELATED WORKS

3. REPLY AND RETWEET USAGE BEFORE AND DURING THE DISASTER

4. REPLY AND RETWEET ON THEFOL-LOWER NETWORK

5. INFORMATION CLASSIFICATION

6. CONCLUSION

 

์ˆœ์œผ๋กœ ์†Œ๊ฐœ ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.

 

0. Abstract

Abstract์—์„œ๋Š” ์—ฐ๊ตฌ๋ชฉ์ ๊ณผ, ๋ฐ์ดํ„ฐ์ˆ˜์ง‘ ๋ฐฉ๋ฒ•, ๊ฒฐ๊ณผ๋ฅผ ๊ฐ„๋žตํžˆ ๋ณด์—ฌ์ฃผ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

 

์—ฐ๊ตฌ๋ชฉ์ : ์žฌ๋‚œ ๋ฐœ์ƒ์‹œ Twitter์—์„œ์˜ ์ •๋ณด ๊ณต์œ  ํ–‰๋™ ๋ถ„์„

๋ฐ์ดํ„ฐ ์ˆ˜์ง‘: ๋™์ผ๋ณธ ๋Œ€์ง€์ง„ ๋ฐœ์ƒ ์ „, ๋ฐœ์ƒ ๋™์•ˆ์˜ ํŠธ์œ— ์ˆ˜์ง‘

 

๊ฒฐ๊ณผ

-Reply ๋ฐ Retweet๊ณผ ๊ฐ™์€ ๊ธฐ๋Šฅ์„ ์ž˜ ์‚ฌ์šฉํ•˜์ง€ ์•Š๋˜ ์‚ฌ์šฉ์ž๋Š” ์žฌ๋‚œ ํ›„์— ์ง€์†์ ์œผ๋กœ ์‚ฌ์šฉํ•˜์ง€ ์•Š์Œ

-Retweets์€ ํŠธ์œ„ํ„ฐ์—์„œ ์ •๋ณด๋ฅผ ๊ณต์œ ํ•˜๋Š” ๋ฐ ์ž˜ ์‚ฌ์šฉ๋จ

-Retweets์€ ์ผ๋ฐ˜ ์‚ฌ์šฉ์ž๊ฐ€ ์ œ๊ณต ํ•œ ์ •๋ณด๋ฅผ ๊ณต์œ ํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋  ๋ฟ ์•„๋‹ˆ๋ผ ๋Œ€์ค‘ ๋งค์ฒด์˜ ์ •๋ณด๋ฅผ ์ „๋‹ฌํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋จ

 

1. INTRODUCTION

 

2011 ๋…„ 3 ์›” 11 ์ผ 14:46์— ๋ฐœ์ƒํ•œ ๋™์ผ๋ณธ ๋Œ€์ง€์ง„ ์ „ํ›„ ๋ฐ์ดํ„ฐ๋ฅผ ์ˆ˜์ง‘

- "์žฌํ•ด ์ „" : 3 ์›” 7 ์ผ๋ถ€ํ„ฐ 10 ์ผ๊นŒ์ง€

- ์žฌํ•ด ํ›„" :3 ์›” 11 ์ผ๋ถ€ํ„ฐ 15 ์ผ๊นŒ์ง€๋กœ ์ •์˜

2,711,473 ๋ช…์˜ ์‚ฌ์šฉ์ž๊ฐ€ ๊ฒŒ์‹œํ•œ 362,435,649๊ฐœ์˜ ํŠธ์œ— ์กด์žฌ

Follower ๋„คํŠธ์›Œํฌ๋กœ ๊ตฌ์„ฑ๋œ ๋ฐ์ดํ„ฐ ์‚ฌ์šฉ

 

2. RELATED WORKS

๋‹ค๋ฅธ ์—ฐ๊ตฌ

์†Œ์…œ ๋ฏธ๋””์–ด๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์‹ค์ œ ์„ธ๊ณ„์—์„œ ์ผ์–ด๋‚œ ์‚ฌ๊ฑด์„ ๊ด€์ฐฐ

Ex) ํŠธ์œ„ํ„ฐ์—์„œ ๋‰ด์Šค์˜ ํผ์ง ์—ฐ๊ตฌ, Twitter๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ง€์ง„ ์ง„์›์ง€๋ฅผ ์ถ”์ •, reply๊ณผ hashtag์— ์ค‘์ ์„ ๋‘์–ด Twitter ์‚ฌ์šฉ์ž ํ–‰๋™์„ ์กฐ์‚ฌ

 

๋ณธ ์—ฐ๊ตฌ

์œ„๊ธฐ ์ƒํ™ฉ์—์„œ ์†Œ์…œ ๋ฏธ๋””์–ด๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์‹ค์ œ ์ƒํ™ฉ์„ ๊ด€์ฐฐ

์ •์ƒ์ ์ธ ์ƒํ™ฉ๊ณผ ์œ„๊ธฐ ์ƒํ™ฉ ์‚ฌ์ด์˜ ์‚ฌ์šฉ์ž ํ–‰๋™ ๋ณ€ํ™”์— ์ง‘์ค‘

 

3. REPLY AND RETWEET USAGE BEFORE AND DURING THE DISASTER

Figure 1: Cumulative rate of retweet users after disaster

X์ถ•์€ ์‹œ๊ฐ„ ์ง€์ง„์ด ๋ฐœ์ƒํ•œ ๋‚ ์งœ๋ถ€ํ„ฐ ์‹œ๊ฐ„์„ ์˜๋ฏธํ•˜๊ณ ,

Y์ถ•์€ ๋ฆฌํŠธ์œ—ํ•œ ์‚ฌ์šฉ์ž๋“ค์˜ ๋ˆ„์  ๋น„์œจ์ž…๋‹ˆ๋‹ค.

3 ์›” 12 ์ผ ๋ง

Pre-retweeters ์˜ 69.0 %๊ฐ€ ์žฌ๋‚œ ํ›„ retweet

๋ฐ˜๋ฉด, Non-retweeter ๋“ค์˜ 21.4%๋งŒ์ด ์žฌ๋‚œ ํ›„ retweet

3 ์›” 23 ์ผ ๋ง

์ „์ฒด ์‚ฌ์šฉ์ž์˜ 52 %์™€ Pre-retweeters ์˜ 90.8 %๊ฐ€ retweet

๊ทธ๋Ÿฌ๋‚˜ Non-retweeter ์˜ 43.1 %๋งŒ์ด retweet

->  ์žฌ๋‚œ ์ด์ „์— ๋ฆฌํŠธ์œ—ํ•˜์ง€ ์•Š์€ ์‚ฌ์šฉ์ž์˜ ์ ˆ๋ฐ˜ ์ด์ƒ์ด ์žฌ๋‚œ ํ›„์—๋„ ๋ฆฌํŠธ์œ—ํ•˜์ง€ ์•Š์Œ

 

 

 

*Pre-retweeters(520,302 tweets)

: ์žฌ๋‚œ ์ „์— ๋ฆฌํŠธ์œ— ๊ธฐ๋Šฅ์„ ์‚ฌ์šฉํ•˜๋˜ ์‚ฌ์šฉ์ž

*Non-retweeter(2,191,171 tweets)

: ์žฌ๋‚œ ์ „์— ๋ฆฌํŠธ์œ—์„ ํ•˜์ง€ ์•Š๋˜ ์‚ฌ์šฉ์ž

 

 

 

4.  REPLY AND RETWEET ON THE FOLLOWER NETWORK

Followers ๋„คํŠธ์›Œํฌ์—์„œ ์ƒํ˜ธ ์ž‘์šฉ์ด ๋ฐœ์ƒํ–ˆ๋Š”์ง€ ์—ฌ๋ถ€์— ๋”ฐ๋ผ Reply๊ณผ Retweet๋กœ ๋ถ„๋ฅ˜

 

Reply์™€ Retweet์˜ ์ •์˜? 

1) Reply

 ์‚ฌ์ „์  ์˜๋ฏธ: "๋Œ€๋‹ต/์‘๋‹ตํ•˜๋‹ค

์ƒ๋Œ€๋ฐฉ์ด ์–ด๋–ค ์–˜๊ธธ ํ•˜๋ฉด ๊ทธ๊ฒƒ์— ๋Œ€ํ•ด ์ž์‹ ์ด ๋ฐ˜์‘์„ ๋ณด์ด๋Š” ๋ง

๋‚˜์™€ ์ƒ๋Œ€๋ฐฉ, ๊ทธ๋ฆฌ๊ณ  ๋‚˜์™€ ์ƒ๋Œ€๋ฐฉ์„ ๋ชจ๋‘ follow ํ•˜๋Š” ์ œ3์ž(๋“ค)๋งŒ ์ž์‹ ์˜ ํƒ€์ž„๋ผ์ธ์—์„œ ๋ณผ ์ˆ˜ ์žˆ์Œ

 

2)Retweet

๋‚จ์ด ์˜ฌ๋ฆฐ ๊ธ€(ํŠธ์œ—, tweet)์„ ๋‹ค์‹œ(re-) ๋‚ด ๊ณ„์ •์„ ํ†ตํ•ด ์˜ฌ๋ฆฐ๋‹ค๋Š” ์˜๋ฏธ

๋‚ด๊ฐ€ ์ ‘ํ•œ ์œ ์šฉํ•œ ์ •๋ณด๋‚˜ ๋‰ด์Šค๋ฅผ ๋‹ค๋ฅธ ์‚ฌ๋žŒ๋“ค(follower)์—๊ฒŒ ์•Œ๋ฆฌ๊ณ ์ž ํ•  ๋•Œ ์‚ฌ์šฉ

Reply ์™€ ๋‹ฌ๋ฆฌ retweet์€ ์•„์ด๋”” ์•ž์— 'RT'๊ฐ€ ๋“ค์–ด๊ฐ์œผ๋กœ์จ ๋‚˜๋ฅผ follow ํ•˜๋Š” ๋ชจ๋“  ์‚ฌ๋žŒ(follower)์—๊ฒŒ ๊ณต๊ฐœ๋จ

(์ถœ์ฒ˜:https://twitteran.tistory.com/entry/difference-reply-and-retweet)

 

Figure 2: Rate of replies and retweets on follower networks

๋จผ์ € ๊ทธ๋ž˜ํ”„์˜ x์ถ•์€ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ์‹œ๊ฐ„์ด๊ณ ,

y์ถ•์€ ํŒ”๋กœ์›Œ ๋„คํŠธ์›Œํฌ์—์„œ์˜ ๊ฐ๊ฐ reply์™€ retweet๋น„์œจ์ž…๋‹ˆ๋‹ค.

 

[Replies]

1)์žฌํ•ด ์ „

- 32%

2)์žฌํ•ด ํ›„

-ํฐ ๋ณ€ํ™” ์—†์Œ

-Reply ์˜ 72.3% ๊ฐ€ ์žฌํ•ด ์ „ ๊ด€๊ณ„๊ฐ€ ์žˆ๋˜ ์‚ฌ๋žŒ์— ๋Œ€ํ•œ reply์˜€์Œ

 

[Retweet]

1)์žฌํ•ด ์ „

-  23%

2)์žฌํ•ด ํ›„

-10% ๋กœ ๊ฐ์†Œ

- ๋ฆฌํŠธ์œ—ํ•˜๋Š” ๊ฒŒ์‹œ๋ฌผ์˜ ์ข…๋ฅ˜๊ฐ€ ํ‰์ƒ์‹œ์™€ ๋‹ค๋ฅด๊ธฐ ๋•Œ๋ฌธ์— ํŒ”๋กœ์›Œ ๊ทธ๋ฃน์•ˆ์—์„œ์˜ ๋ฆฌํŠธ์œ—์€ ๊ฐ์†Œ

- ํŒ”๋กœ์›Œ ๋„คํŠธ์›Œํฌ ๋ฐ–์—์„œ์˜ ๋ฆฌํŠธ์œ—์€ ์ฆ๊ฐ€

 

 

5. INFORMATION CLASSIFICATION

ํŠธ์œ„ํ„ฐ์—์„œ ๊ด‘๋ฒ”์œ„ํ•˜๊ฒŒ ์‚ฌ์šฉ๋œ ์ •๋ณด๋ฅผ ๋ถ„๋ฅ˜ํ•˜์—ฌ ์‚ฌ๋žŒ๋“ค์ด ์žฌ๋‚œ ์ค‘์— ์–ด๋–ค ์ข…๋ฅ˜์˜ ์ •๋ณด๋ฅผ ์š”๊ตฌํ•˜๋Š”์ง€ ์—ฐ๊ตฌ

 

5.1 Retweet Clustering

1000 ๋ฒˆ ์ด์ƒ Retweet ๋œ ํŠธ์œ—์„ ์‚ฌ์šฉ

Ui =  ๋ฆฌํŠธ์œ— i๋ฅผ ๋ฆฌํŠธ์œ—ํ•œ ์œ ์ €๋“ค์˜ ๊ทธ๋ฃน

Uj =  ๋ฆฌํŠธ์œ— j๋ฅผ ๋ฆฌํŠธ์œ—ํ•œ ์œ ์ €๋“ค์˜ ๊ทธ๋ฃน

Oij = ๊ทธ๋ฃนUi์™€ ๊ทธ๋ฃนUj์— ์œ ์ €๋“ค์ด ๊ฒน์น˜๋Š” ์ •๋„( Jaccard co-efficient ์ ์šฉ)

๋™์ผํ•œ ์œ ์ €์— ์˜ํ•ด Retweet ๋œ ํŠธ์œ—์€ ์œ ์‚ฌํ•œ ํŠน์ง•์„ ๊ฐ€์ง€๊ณ  ์žˆ์Œ์„ ์˜๋ฏธ

  -> ์ด๋Ÿฌํ•œ ๊ด€๊ณ„๋ฅผ ๊ฐ€์ง„ ํŠธ์œ— ์Œ๋“ค์€ ๋„คํŠธ์›Œํฌ๋ฅผ ์ƒ์„ฑ

์กฐ๊ฑด  : Oij๊ฐ€ ์ž„๊ณ„์น˜๋ฅผ ๋„˜๋Š” ๋ฆฌํŠธ์œ—์ด์–ด์•ผํ•จ(th = 0.04 ๋กœ ์„ค์ •)

 

 

Jaccard co-efficient ์ด๋ž€?

๋‘ ์ง‘ํ•ฉ ์‚ฌ์ด์˜ ์œ ์‚ฌ๋„๋ฅผ ์ธก์ •ํ•˜๋Š” ๋ฐฉ๋ฒ• ์ค‘ ํ•˜๋‚˜

0๊ณผ 1 ์‚ฌ์ด์˜ ๊ฐ’์„ ๊ฐ€์ง

๋‘ ์ง‘ํ•ฉ์ด ๋™์ผํ•˜๋ฉด 1, ๊ณตํ†ต์˜ ์›์†Œ๊ฐ€ ํ•˜๋‚˜๋„ ์—†์œผ๋ฉด 0

 

์ถœ์ฒ˜:https://ko.wikipedia.org/wiki/%EC%9E%90%EC%B9%B4%EB%93%9C_%EC%A7%80%EC%88%98



5.2 Types of Diffused Information

Figure 3: Retweet network

 

๊ทธ๋ฆผ์—์„œ ๋…ธ๋“œ๋“ค์€ ๊ฐ ๋ฆฌํŠธ์œ—์„ ๋งํ•˜๊ณ , ์—ฃ์ง€๋“ค์€ ๋™์ผํ•œ ์œ ์ €๊ฐ€ ๋ฆฌํŠธ์œ—์„ ํ•œ ์ •๋„๊ฐ€ 0.4๋ฅผ ๋„˜์—ˆ์„ ๊ฒฝ์šฐ ๋„คํŠธ์›Œํฌ๋ฅผ ํ˜•์„ฑํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค.

 

๊ทธ ์™€์ค‘์— ํŠน์ง•์ ์ธ ๋„คํŠธ์›Œํฌ๋“ค์ด ๋ณด์ด๋Š”๋ฐ

์ด ํŠธ์œ—๋“ค์˜ ๊ณตํ†ต์ ์„ ๋ถ„์„ํ•ด๋ณด๋‹ˆ, ์•„๋ž˜์™€ ๊ฐ™์€ ์ด๋Ÿฌํ•œ ํŠน์ง•์ด ์žˆ์—ˆ๊ณ ,

Table 1: Spread information by retweets

๊ฐ€์žฅ ํฐ ๊ตฌ์„ฑ์š”์†Œ๋ฅผ ๋ถ„์„ํ•ด๋ณธ ๊ฒฐ๊ณผ, ๋‹ค์Œ๊ณผ ๊ฐ™์•˜์Šต๋‹ˆ๋‹ค.

Top ๏ฌve largest components

A : ๋Œ€์ค‘ ๋งค์ฒด์™€ ๊ณต๊ณต ๊ณ„์ •์˜ ๊ด€์‹ฌ์„ ๋„๋Š” ์ •๋ณด

B : ๋น„๊ณต๊ฐœ ์‚ฌ์šฉ์ž๊ฐ€ ํŠธ์œ— ํ•œ ์ง€์ง„์˜ ์ฃผ์˜๋ฅผ ํ™˜๊ธฐ์‹œํ‚ค๋Š” ๋Œ€๋Ÿ‰ ์ •๋ณด

C : ํ›„์ฟ ์‹œ๋งˆ ์›์ž๋ ฅ ๋ฐœ์ „์†Œ ์‚ฌ๊ณ ์— ๊ด€ํ•œ ์œ ํ•ดํ•œ ์†Œ๋ฌธ์— ๋Œ€ํ•ด ๊ฒฝ๊ณ 

D : ๋Œ€์ค‘ ๋งค์ฒด ๋ฐ ์ง€๋ฐฉ ์ •๋ถ€์™€ ๊ฐ™์€ ๊ณต๊ณต ๊ณ„์ •์œผ๋กœ ํŠธ์œ— ๋œ ๊ณ„ํš๋œ ์ •์ „์— ๋Œ€ํ•œ ์ •๋ณด

E  : ์›์ž๋ ฅ ๋ฐœ์ „์†Œ ์žฌํ•ด ๋ฐ ๋ฐฉ์‚ฌ๋Šฅ์— ๋Œ€ํ•œ ์ •๋ณด

 

6. CONCLUSION

๋™์ผ๋ณธ ๋Œ€์ง€์ง„ ์ „ํ›„์— ๊ฒŒ์‹œ ๋œ 3 ์–ต 6 ์ฒœ๋งŒ ๊ฐœ์˜ ํŠธ์œ—์„ ๋ถ„์„

์žฌ๋‚œ ๋ฐœ์ƒ์‹œ ์‚ฌ๋žŒ๋“ค์ด Twitter์—์„œ ์ •๋ณด๋ฅผ ๊ณต์œ ํ•˜๋Š” ๋ฐฉ์‹์„ ์„ค๋ช…

 

๊ตฌ์ฒด์ ์ธ ๊ฒฐ๋ก 

-์žฌ๋‚œ ํ›„์—๋Š” Retweet์ด Twitter์—์„œ ์ •๋ณด๋ฅผ ๊ณต์œ ํ•˜๋Š” ๋ฐ ์ž˜ ์‚ฌ์šฉ๋จ

-์ผ๋ฐ˜ ์‚ฌ์šฉ์ž๊ฐ€ ์ œ๊ณตํ•œ ์ •๋ณด๋ฅผ ๊ณต์œ  ํ• ๋ฟ ์•„๋‹ˆ๋ผ ๋Œ€์ค‘ ๋งค์ฒด์˜ ์ •๋ณด๋ฅผ ์ „๋‹ฌํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ