CrowdsouRS is a crowdsourced reputation system presented as a browser extension that allows users to rate the trustworthiness of websites on a 1 to 5 scale. The extension communicates ratings to a centralized server which calculates reputation scores for websites using a Bayesian method. A user study found the system was effective at identifying deceptive websites and most users believed CrowdsouRS could help address misleading online content. However, limitations included a small number of ratings for some sites and a biased participant pool.
Crowdsourced Reputation System Identifies Deceptive Web Content
1. CrowdsouRS: A Crowdsourced Reputation
System for
Identifying Deceptive Web-contents
Presented By
Md. AbuTalha Masiur Rahman Siddiki
Reg: 2012331008 Reg: 2012331028
CSE, SUST CSE, SUST
Supervised By
Dr. Farida Chowdhury
Assistant Professor
Dept. of CSE, SUST
5. Motivation (Cont.)
▪ From Hate SpeechTo Fake News:The Content Crisis Facing Mark
Zuckerberg.[1]
▪ Facebook, GoogleTake StepsTo Confront Fake News.[2]
▪ Google steps up fight against fake news by expanding fact-check tool.
[3]
6. Motivation (Cont.)
▪ Poorly preparation for controlling and sanctioning the substantial
and increasing number of users and service providers with unethical,
malicious and criminal intentions.
7. Motivation (Cont.)
▪ Internet activities are not uniformly well intentioned, because they
are increasingly motivated by financial profit and personal gain which
can lead to unethical and criminal behavior.
▪ Early optimism of Internet replaced by cynicism and diminishing
trust in the Internet as a reliable platform for building markets and
communities.
▪ The purpose of security mechanisms is to provide protection against
malicious parties.
▪ Soft security mechanisms that can provide protection based on input
from the whole community.
8. Our Contribution
We present the CrowdsouRS, a browser-extension based
Trust and Reputation system for the web.
We present its architecture and implementation details
to illustrate our design choices that make it easily
deployable.
We present a thorough user-study to test the usability
and applicability of our system.
9. What is Trust?
▪ Trust, according to Deutsch (1973) is:
“Confidence that [one] will find what is
desired [from another] rather than what is feared.”
▪ Trust is a directional relationship between
two parties that can be called trustor and
trustee.
10. What is Reputation?
Reputation is about character or standing. A strong reputation builds
trust.
According to Cambridge English Dictionary,
"The opinion that people in general have about someone or something, or
how much respect or admiration someone or something receives, based on
past behavior or character.“
11. Trust Vs Reputation
▪ According to Josang, differences between trust and reputation can
be illustrated with the following statements:
a. "I trust Hanif Enterprise because of their reputation."
b. "I trust EnaTransport in spite of their bad reputation."
12. Reputation System
▪ A reputation system computes scores or ratings from community or
each consumer or a system called reputation computation engine.
Two unknown parties can be beneficiary from this system.
Consumer can take the decision looking at reputation
Service provider can also sell its service or product.
13. Reputation System (Cont.)
According to Resnick et al. reputation systems must have the following three
properties to operate at all:
▪ Longevity factor should be taken into account so that it can lead to an
expectation of future interactions.
▪ Current interactions ratings are grabbed and shared.
▪ Decisions about current interactions should be on the basis of ratings about
previous interactions.
16. Methodology
▪ We need to collect data from communities, users .
▪ For reputation system implementation in a general level we describe
a simple trust & reputation system extension named “CrowdsouRS”
which can be installed on browser.
18. Why Extension?
▪ This allows the reputation score of any Web page to be visualized to
the user, as well as the user to rateWeb sites andWeb pages.
▪ The extension communicates with a centralized server which keeps
the reputation scores of allWeb pages.
▪ AWeb page can be rated by the user with a discrete set of different
levels as described above.This architecture is illustrated in next.
24. Collecting Data
From browser’s active tab’s URL data is collected on 1 to 5 scale. The
scale is described as follows:
– Very Untrustworthy
– Untrustworthy
– Average
– Trustworthy
– VeryTrustworthy
25. Method Selection
▪ We have used the Bayesian system which encompasses
Bayesian multinomial system.
▪ The Bayesian multinomial system mainly is based on Dirichlet
probability distribution.
▪ As we have used one to five scale rating system, this method is
appropriate for our study.
26. Why Bayesian multinomial system ?
▪ Other methods vary so hugely, and they do not take longevity into
account.
▪ Bayesian can reduces loss function.
▪ Bayesian is more stable and robust method to compute reputation
score.
27. Score Calculation (Cont.)
Now, we use the formula
Here,
– i represents rating levels,
– sy(i) represents the score of I,
– Ry(i) represents the total number of votes of I,
– C is the value of choice , we consider c = 2 here,
– a(i) represents base rate of i.
28. Score Calculation (Cont.)
Now, Our actual reputation score can be computed as:
As we are taking five different levels i.e k = 5, v(i) can be:
30. Result Analysis(Cont.)
The result of average reputation scores for different websites as
generated by CrowdsouRS utilizing the ratings provided by the
participants is presented in previous slide.
– According to this figure, more than 70% people assume that Facebook and Google are
very trustworthy while only less than 2% people mark these web sites untrustworthy.
– However, when we calculate the reputation score of abcnews.com.co, 75% people
rightfully indicate that this website contains deceptive misleading contents.
This shows that CrowdsouRS has been able to identify deceptive
contents which validates our assumption.
32. Satisfaction with the Reliability of CrowdsouRS
We have evaluated the experience and
perception of users regarding CrowdsouRS.
It is evident from the result that the user
experience has been mostly positive.
Among all users, around 5%
participants have been satisfied.
And 22%of the participants have
been highly satisfied.
While 29.3% participants remain
neutral.
33. Influence of CrowdsouRS
Most of the participants
(around 70%) believe that
CrowdsouRS can be an
effective tool to fight against
deceptive misleading online
contents.
34. Limitations
Many problems and drawbacks in the reputation system and our study is
not free of limitations.
▪ The number of votes cast in the database is minimal in amount for
some websites and URLs.
▪ The majority of our participants from the campus-oriented
neighborhood (95% were students, and others were recent ex-
students) that our participant pool had been biased, they did not
represent the general public.
35. References
[1] NPR.org. (2016). From Hate SpeechTo Fake News:The Content Crisis Facing Mark
Zuckerberg. [online]Available at:
http://www.npr.org/sections/alltechconsidered/2016/11/17/495827410/from-hate-
speech-to-fake-news-the-content-crisis-facing-mark-zuckerberg [Accessed 18 Apr.
2017].
[2] NPR.org. (2016). Facebook,GoogleTake StepsTo Confront Fake News. [online]
Available at:
http://www.npr.org/sections/alltechconsidered/2016/11/15/502111390/facebook-
google-take-steps-to-confront-fake-news [Accessed 18 Apr. 2017].
[3] Sulleyman,A. (2017). Google steps up fight against fake news by expanding fact-check
tool. [online]The Independent. Available at: http://www.independent.co.uk/life-
style/gadgets-and-tech/news/google-fake-news-fight-fact-checking-tool-expand-
three-countries-brazil-mexico-argentina-a7589741.html [Accessed 18Apr. 2017].