answering twitter questions: a model for recommending...

24
Answering Twitter Questions: a Model for Recommending Answerers through Social Collaboration DATE: 2017/05/02 AUTHOR: Laure Soulier, Lynda Tamine, Gia-Hung Nguyen SOURCE: CIKM’16 ADVISOR: Jia-Ling, Koh SPEAKER: Pei-Hao, Wu 1

Upload: others

Post on 22-Jul-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Answering Twitter Questions: a Model for Recommending Answerers through Social Collaboration

DATE: 2017/05/02

AUTHOR: Laure Soul ier, Lynda Tamine, G ia-Hung Nguyen

SOURCE: CIKM’16

ADVISOR: J ia -L ing , Koh

SPEAKER: Pe i -Hao, Wu

1

Page 2: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Outline

•Introduction

•Method

•Experiment

•Conclusion

2

Page 3: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Introduction◦ Motivation

3

Page 4: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Introduction◦ Problem

Majority of questions do not receive a reply

Minority receives a high number of responses

4

Page 5: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Introduction◦ Goal

◦ Answering Twitter questions through social collaboration

5

Page 6: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Introduction◦ Flow chart

Question twitter

6

Users:Authority-base feature, Feedback feature,Complementarity-based feature

Answer twitter

Page 7: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Outline

•Introduction

•Method

•Experiment

•Conclusion

7

Page 8: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method◦ Collaboration likelihood model

◦ Feature

◦ (1) whether the mentioned user 𝑢𝑗′ provides feedback to user 𝑢𝑗 (reply, retweet ,mention)

◦ (2) authority-based features

◦ (3) complementarity-based features

◦ Collaboration likelihood

◦ :𝑢𝑗 and 𝑢𝑗′ estimate value

8

Page 9: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method◦ IG(g, 𝑢𝑘) : group-based information gain

◦ H(g) : group entropy

◦ H(g|𝑢𝑘) : conditional entropy

9

Page 10: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method

10

𝑢1 𝑢2 𝑢3 𝑢4

qHow do you get infected by this Ebola virus though?

1 1

1

21

3

H(g) = -(1

2log

1

2) -(

1

3log

1

3) -(1 log1) -(1 log1)

= 1

2log2 +

1

3log3 + 0 + 0

= 1.028

Page 11: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method

11

H(g|𝑢1)

= 1 * [ -(1

2log

1

2) -(

1

2log

1

2) -(0 log0)]

= 1

2log2 +

1

2log2 + 0

= log2= 1

1

2

𝑢1

𝑢2

𝑢3

𝑢4

1

2

0

1

3

𝑢1

𝑢2

𝑢3

𝑢41

3

1

3

H(g|𝑢2)

= 1

2* [ 1.584]

= 0.792

Page 12: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method

12

1

4𝑢1

𝑢2

𝑢3

𝑢4

1

2

1

4

1

5

𝑢1

𝑢2

𝑢3

𝑢43

5

1

5

H(g|𝑢3)

= 1

3* [ 1.5]

= 0.5

H(g|𝑢4)

= 1

5* [ 1.37]

= 0.274

Page 13: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method◦ Building the recommended collaborative group

13

H(g) = 1.028

IG(g|𝑢1) = 1.028 – 1 = 0.028

IG(g|𝑢2) = 1.028 – 0.792 = 0.236

IG(g|𝑢3) = 1.028 – 0.5 = 0.528

IG(g|𝑢4) = 1.028 – 0.274 = 0.754

0.028𝑢1

𝑢2

𝑢3

𝑢4

Round 1:

0.236

0.528

0.754

Page 14: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method◦ Building the recommended collaborative group

14

H(g) = 1

3log3 + 0 + 0 = 0.528

IG(g|𝑢2) = 0.528 – 0.792 = -0.264

IG(g|𝑢3) = 0.528 – 0.5 = 0.028

IG(g|𝑢4) = 0.528 – 0.274 = 0.254 𝑢2

𝑢3

𝑢4

Round 2:

0.236

0.528

0.754

Page 15: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method◦ Building the recommended collaborative group

15

H(g) = 0 + 0 = 0

IG(g|𝑢3) = 0 – 0.5 = - 0.5

IG(g|𝑢4) = 0 – 0.274 = - 0.254

𝑢3

𝑢4

Final round:

0.528

0.754

Page 16: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Method◦ Recommended answers by the users of collaborative group

16

𝑢3

𝑢4

Answer twitter

You can get Ebola though Food

By eating infected bats, monkeys, contaminated food

qHow do you get infected by this Ebola virus though?

Answer the question

Page 17: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Outline

•Introduction

•Method

•Experiment

•Conclusion

17

Page 18: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Experiment◦ Datasets

◦ Data from two subjects

◦ Sandy hurricane

◦ 4,853,345 English tweets

◦ 2012.10.29 – 2012.10.31

◦ Ebola

◦ 4,815,142 English tweets

◦ 2014.8.28 – 2012.08.29

18

Page 19: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Experiment◦ Methods compared

19

Page 20: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Experiment◦ Methods compared

20

Page 21: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Experiment◦ Parameter Sensitivity

21

Page 22: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Experiment◦ Case Study

22

Page 23: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Outline

•Introduction

•Method

•Experiment

•Conclusion

23

Page 24: Answering Twitter Questions: a Model for Recommending ...184pc128.csie.ntnu.edu.tw/presentation/17-05-02...Answering Twitter Questions: a Model for Recommending Answerers through Social

Conclusion◦ Build predictive model to characterize the collaboration likelihood

between pairwise users who are collectively able to solve target tweeted questions

◦ Recommending a collaborative group of users to answer the tweeted question

24