how opinions are received by online communities a case study on amazon.com helpfulness votes...
TRANSCRIPT
How Opinions are Received by Online Communi-tiesA Case Study on Amazon.com Helpfulness Votes
Cristian Danescu-Niculescu-Mizil1, Gueorgi Kossinets2, Jon Kleinberg1, Lillian Lee1
1Dept. of Computer Science, Cornell University, 2Google Inc.
WWW 2009
2009. 07. 30.
IDS Lab.
Hwang Inbeom
Copyright 2009 by CEBT
Outline
Users’ evaluation on online reviews: Helpfulness votes
Observation of behaviors
Making some hypothesis and proving their validity
Coming up with a mathematical model explains these be-haviors
2
Copyright 2009 by CEBT
Introduction
Opinion
What did Y think of X?
3
Copyright 2009 by CEBT
Introduction
Meta-Opinion
What did Z think of Y’s opinion of X?
4
Copyright 2009 by CEBT
The Helpfulness of Reviews
Widely-used web sites include not just reviews, but also evaluations of the helpfulness of the reviews
The helpfulness vote
– “Was this review helpful to you?”
Helpfulness ratio:
– “a out of b people found the review itself helpful”
5
b
a
Copyright 2009 by CEBT
Amazon.com Helpfulness Votes Data
4,000,000 reviews about roughly 700,000 books, includ-ing average star ratings and helpfulness ratios
6
Average star rating
Helpfulness ratio
Copyright 2009 by CEBT
Definitions of “Helpfulness”
Helpfulness in the narrow sense: “Does this review help you in making a purchase
decision?”
Liu’s work: annotation and classification of review helpful-ness
Annotators’ evaluation differed significantly from the help-fulness votes
Helpfulness “in the wild”
The way Amazon users evaluate each others’ reviews
Intertwined with complex social feedback mechanisms
7
Copyright 2009 by CEBT
Flow of Presentation
Hypothe-siz-ing
Verify-ing
Model-ing
8
Copyright 2009 by CEBT
Flow of Presentation
Hypoth-esizing•Con-formity
•Indi-vidual-bias
•Bril-liant-but-cruel
•Qual-ity-only
Verifying Modeling
9
Copyright 2009 by CEBT
Hypotheses: Social Mechanisms underlying
Well-studied hypotheses for how social effects influence group’s reaction to an opinion
The conformity hypothesis
The individual-bias hypothesis
The brilliant-but-cruel hypothesis
The quality-only straw-man hypothesis
10
Copyright 2009 by CEBT
Hypotheses
The conformity hypothesis
Review is evaluated as more helpful when its star rating is closer to the consensus star rating
– Helpfulness ratio will be the highest of which reviews have star rating equal to overall average
The individual-bias hypothesis
When a user considers a review, he or she will rate it more highly if it expresses an opinion that he or she agrees with
11
Copyright 2009 by CEBT
Hypotheses (contd.)
The brilliant-but-cruel hypothesis
Negative reviewers are perceived as more intelligent, com-petent, and expert than positive reviewers
The Quality-only straw-man hypothesis
Helpfulness is being evaluated purely based on the textual content of reviews
Non-textual factors are simply correlates of textual quality
12
Copyright 2009 by CEBT
Flow of Presentation
Hypothe-siz-ing
Verifying•Absolute deviation of helpful-ness ratio
•Signed deviation of helpful-ness ratio
•Variance of star rat-ing and helpful-ness ratio
•Making use of plagiarism
Model-ing
13
Copyright 2009 by CEBT
Hypotheses
Conformity•A review is evaluated as more helpful when its star rating is closer to the average star rating
Individual-bias•A review is evaluated as more helpful when its star rating is closer to evaluator’s opinion
Brilliant-but-cruel•A review is evaluated as more helpful when its star rating is below to the average star rating
Quality-only•Only textual infor-mation affects help-fulness evaluation
14
Copyright 2009 by CEBT
Absolute Deviation from Average
Consistent with conformity hypothesis
Strong inverse correlation between the median helpfulness ratio and the absolute deviation
Reviews with star rating close to the average gets higher helpfulness ratio
15
Copyright 2009 by CEBT
Hypotheses
Conformity•A review is evaluated as more helpful when its star rating is closer to the average star rating
Individual-bias•A review is evaluated as more helpful when its star rating is closer to evaluator’s opinion
Brilliant-but-cruel•A review is evaluated as more helpful when its star rating is below to the average star rating
Quality-only•Only textual infor-mation affects help-fulness evaluation
16
Copyright 2009 by CEBT
Signed Deviation from Average
Not consistent with brilliant-but-cruel hypothesis
There is tendency towards positivity
Black lines should not be sloped that way if it is valid hy-pothesis
17
Copyright 2009 by CEBT
Hypotheses
Conformity•A review is evaluated as more helpful when its star rating is closer to the average star rating
Individual-bias•A review is evaluated as more helpful when its star rating is closer to evaluator’s opinion
Brilliant-but-cruel•A review is evaluated as more helpful when its star rating is below to the average star rating
Quality-only•Only textual infor-mation affects help-fulness evaluation
18
Copyright 2009 by CEBT
Addressing Individual-bias Effects
It is hard to distinguish between the conformity and the individual-bias hypothesis
We need to examine cases in which individual people’s opinions do not come from exactly the same distribution
Cases in which there is high variance in star ratings
Otherwise conformity and individual-bias are indistinguish-able
– Everyone has same opinion
19
Copyright 2009 by CEBT
Variance of Star Rating and Helpfulness Ra-tio
20
Helpfulness ratio is the highest with re-views of which rating is slightly-above the average
Two-humped camel plots: local minimum around average
Helpfulness ratio is the highest when star ratings of re-views have average value
Copyright 2009 by CEBT
Hypotheses
Conformity•A review is evaluated as more helpful when its star rating is closer to the average star rating
Individual-bias•A review is evaluated as more helpful when its star rating is closer to evaluator’s opinion
Brilliant-but-cruel•A review is evaluated as more helpful when its star rating is below to the average star rating
Quality-only•Only textual infor-mation affects help-fulness evaluation
21
Copyright 2009 by CEBT
Plagiarism
Making use of plagiarism is effective way to control for the effect of review text
Definition of plagiarized pair(s) of reviews
Two or more reviews of different products
With near-complete textual overlap
22
Copyright 2009 by CEBT
An Example
Skull-splitting headache guaranteed!•If you enjoy thumping, skull splitting migraine headache, then Sing N Learn is for you.As a longtime language instructor, I agree with the attempt and effort that this series makes, but it is the execution that ultimately weakens Sing N Learn Chinese.To be sure, there are much, much better ways to learn Chi-nese. In fact, I would recommend this title only as a last re-sort and after you’ve thoroughly exhausted traditional ways to learn Chinese …
Migraine Headache at No Extra Charge•If you enjoy a thumping, skull splitting migraine headache, then the Sing N Learn series is for you.As a longtime language instructor, I agree with the effort that this series makes, but it is the execution that ultimately weakens Sing N Learn series. To be sure, there are much, much better ways to learn a foreign language. In fact, I would recommend this title only as a last resort and after you’ve thoroughly exhausted traditional ways to learn Korean …
23
Copyright 2009 by CEBT
Plagiarism (contd.)
Plagiarized reviews
Almost(not exact) same text
– More possibly, same text could be considered as spam reviews
Different non-textual information
If the quality-only straw man hypothesis holds, helpful-ness ratios of documents in each pair should be the same
Possible other methods
Human annotation
– Could be subjective
Classification using machine learning methods
– We cannot guarantee the accuracies of algorithms
24
Copyright 2009 by CEBT
Experiments with Plagiarism
Text quality is not the only explanatory factor
Statistically significant difference between the helpfulness ratios of plagiarized pairs
25
The plagiarized reviews with deviation 1 is significantly more helpful than those with deviation 1.5
Copyright 2009 by CEBT
Hypotheses
Conformity•A review is evaluated as more helpful when its star rating is closer to the average star rating
Individual-bias•A review is evaluated as more helpful when its star rating is closer to evaluator’s opinion
Brilliant-but-cruel•A review is evaluated as more helpful when its star rating is below to the average star rating
Quality-only•Only textual infor-mation affects help-fulness evaluation
26
Copyright 2009 by CEBT
Flow of Presentation
Hypothe-siz-ing
Verify-ing
Model-ing•Based on in-divid-ual bias andmix-tures of dis-tribu-tions
27
Copyright 2009 by CEBT
Authors’ Model
Based on individual bias and mixtures of distributions
Two distributions: one for positive, one for negative eval-uators
Balance between positive and negative evaluators:
Controversy level:
– Density function of helpfulness ratios of positive evaluators
– Gaussian distribution of which average is -centered
– Density function of helpfulness ratios of negative evaluators
– Gaussian distribution of which average is -centered
28
)( q
)( p
)()()( xqgxpfxh
)(xg
)(xf
p
Copyright 2009 by CEBT
Validity of the Model
Empirical observation and model generated
29
Copyright 2009 by CEBT
Conclusion
A review’s perceived helpfulness depends not just on its content, but also the relation of its score to other scores
The dependence of the score is consistent with a simple and natural model of individual-bias in the presence of a mixture of opinion distributions
Directions for further research
Variations in the effect can be used to form hypotheses about differences in the collective behaviors of the underly-ing populations
It would be interesting to consider social feedback mecha-nisms that might be capable of modifying the effects au-thors observed here
Considering possible outcomes of design problem for sys-tems enabling the expression and dissemination of opinions
30
Copyright 2009 by CEBT
Discussions
So, how can we use this?
In which cases would this information be helpful?
Available information is very limited
– Star ratings
– Helpfulness ratios
Conclusion is rather trivial
Does not present new discoveries
31