data mining paper presentation

44
From Bias to Opinion: a Transfer- Learning Approach to Real-Time Sentiment Analysis Pedro , Adriano, Wagner, Virgilio Universidade Federal de Minas Gerais, Brazil 4/10/2012 Presentation for Comp722 Data Mining, Kaiwen Qi 06/22/2022 1

Upload: kaiwen-qi

Post on 25-May-2015

907 views

Category:

Technology


1 download

TRANSCRIPT

Page 1: Data mining paper presentation

04/12/2023 1

From Bias to Opinion: a Transfer-Learning Approach to Real-Time Sentiment Analysis

Pedro , Adriano, Wagner, Virgilio Universidade Federal de Minas Gerais, Brazil

4/10/2012 Presentation for Comp722 Data Mining, Kaiwen Qi

Page 2: Data mining paper presentation

04/12/2023 2

Outline

Background and Paper Purpose Quantify Bias Exploiting Bias for Sentiment Analysis Conclusions

Page 3: Data mining paper presentation

04/12/2023 3

Social Media and Opinionated Data

From Pedro’s PPT

Page 4: Data mining paper presentation

04/12/2023 4

Background: Sentimental Analysis

GoalDetermine the attitude of a speaker or a writer with respect to some topic or the overall contextual polarity of a document.

Page 5: Data mining paper presentation

04/12/2023 5

Sentimental Analysis

Example: http://www.tweetsentiments.com/analyze?utf8=%E2%9C%93&q=Lady+Gaga&topic=false&commit=Analyze+Tweets

Page 6: Data mining paper presentation

04/12/2023 6

Sentimental Analysis

Another name: Opinion Mining

Page 7: Data mining paper presentation

04/12/2023 7

Use sentimental analysis for:

Help companies keep on top of issues and respond to trends impacting on business.

Gather new customer insights from unstructured-content (gathered from social networks). 

Determine the degree to which a sentiment is positive, negative or neutral for the entire content or a segment of the content.

Identify those voices and publications influencing customers and competitors. Adjust and optimize communication strategies. Use it to direct strategic decisions such as modifying marketing messages,

customer service or product development. Receive early warnings of market developments. Manage and preserve brand equations and reputations. Monitor public opinion Summarize the aggregated sentiment of online society

http://passionjunkie.hubpages.com/hub/Sentimental-Analysis-Business-Insights-that-Help-you-Grow

Page 8: Data mining paper presentation

04/12/2023 8

Real-Time v.s Traditional Sentiment Analysis

Traditional: Uses static and well-controlled scenarios that target analysis of

reviews of products and services Pre-defined Lists of positive and negative words

Real-Time: Lack of labeled textual data Dynamicity of discussion : dynamic/concept drift/non-stationary

distribution

Page 9: Data mining paper presentation

04/12/2023 9

Dynamic Discussion and lack of labeled data

From Pedro’s ppt

Page 10: Data mining paper presentation

04/12/2023 10

Task

What is the time-invariant pattern that does not require significant labeling efforts and supports real-time sentiment analysis?

Page 11: Data mining paper presentation

04/12/2023 11

Proposal

Page 12: Data mining paper presentation

04/12/2023 12

Social Media Endorsements as Evidence of User Bias

Endorsements : Interactions among users in which one user implicitly agrees with another.

Page 13: Data mining paper presentation

04/12/2023 13

Bias and opinions

From Pefro’s PPT

Page 14: Data mining paper presentation

04/12/2023 14

Proposal intension

How can the sociological definition of bias be implemented into a social media platform by only considering social interactions among users?

How can bias information be converted into information on the sentiment that is associated with the generated content?

Page 15: Data mining paper presentation

04/12/2023 15

Modeling User Bias Prediction

Determine the most similar users based on individual endorsements

Page 16: Data mining paper presentation

04/12/2023 16

Measuring bias

We label users whose bias is clearly identifiable as representative of a particular side in a discussion

From Pefro’s PPT

Page 17: Data mining paper presentation

04/12/2023 17

Modeling User Bias Prediction

Activity similarity The similarity considering the users that both

pair of users retweeted

Passive similarity The similarity considering the users that

retweeted both pair of users

Page 18: Data mining paper presentation

18

The Opinion Agreement Graph G=(V,E)

Vertices : User Edge: global judgment of the connected users

04/12/2023

Page 19: Data mining paper presentation

19

The Opinion Agreement Graph

04/12/2023

From Pefro’s PPT

Page 20: Data mining paper presentation

04/12/2023 20

Explanation

Page 21: Data mining paper presentation

04/12/2023 21

Measure Bias

Attractors: sever as reliable sources of bias knowledge

The bias of each node is its proximity from attractors that represent that side to all users in U

Random walk: to measure proximity among nodes

Page 22: Data mining paper presentation

04/12/2023 22

Bias measurement

Page 23: Data mining paper presentation

04/12/2023 23

Case Study

Brazilian 2010 Presidential Elections Brazilian 2010 Soccer League

Page 24: Data mining paper presentation

04/12/2023 24

Bias in Elections Discussions

Page 25: Data mining paper presentation

04/12/2023 25

Bias in Elections Discussions

04/12/2023 25

Page 26: Data mining paper presentation

04/12/2023 26

Bias in Soccer Discussions

Page 27: Data mining paper presentation

04/12/2023 27

Bias in Soccer Discussions

Page 28: Data mining paper presentation

28

Bias is a consistent pattern

From Pedro’s PPT

04/12/2023

Page 29: Data mining paper presentation

04/12/2023 29

Consistent Bias

Page 30: Data mining paper presentation

04/12/2023 30

Background: Transfer Learning

Using learned knowledge from one context to benefit further learning tasks in other contexts

Benefit from knowledge Obtained from similar Tasks or domains

From Liyuan Dai’s paper

Page 31: Data mining paper presentation

04/12/2023 31

Transfer Learning

Example:

Page 32: Data mining paper presentation

04/12/2023 32

Transferring bias from user to content

From Pedro’s PPT

Page 33: Data mining paper presentation

04/12/2023 33

Relationship between terms and users bias

From Pefro’s PPT

Page 34: Data mining paper presentation

04/12/2023 34

Relationship between terms and users bias

From Pedro’s PPT

Page 35: Data mining paper presentation

35

Relationship between terms and users bias

From Pedro’s PPT

04/12/2023

Page 36: Data mining paper presentation

04/12/2023 36

Message Polarity Determination

The term of highest polarity in each tweet:polarity = argmax(p ҄(polarity = x|t))

Page 37: Data mining paper presentation

04/12/2023 37

Evaluating the Knowledge Transfer Process

F1 accuracy v.s number of user with bias When the bias of 15% of users commenting on politics is known, F1=85%

Page 38: Data mining paper presentation

04/12/2023 38

Evaluating the Knowledge Transfer Process

F1 v.s number of users with bias When the bias of 15% of users commenting on politics is known, F1=90%

Page 39: Data mining paper presentation

04/12/2023 39

Comparison to SVM SVM F1 decreases due to the textual feature distribution Bias-based is better, not using labeled textual data Maintain a stable F1, as it incrementally incorporate bias

information on new terms by propagating user bias.

Page 40: Data mining paper presentation

04/12/2023 40

Comparison to SVM SVM F1 decreases Bias-based = SVM, but not require labeled textual data

Page 41: Data mining paper presentation

04/12/2023 41

Analyzing a Soccer Math in Real Time

Live event

From Pedro’s PPT

Page 42: Data mining paper presentation

04/12/2023 42

conclusions Real-time sentiment analysis based on the consistency

of the user bias Known bias Propagate through endorsements propagate user bias to terms associated with user content combine term bias to computer the overall content polarity

Page 43: Data mining paper presentation

04/12/2023 43

Thanks &

Question?

Page 44: Data mining paper presentation

04/12/2023 44

Extra Slides

http://www.cs.cornell.edu/people/pabo/movie-review-data/