Download Amazon - E-Commerce - Lecture Slides and more Slides Fundamentals of E-Commerce in PDF only on Docsity!
AmazonAmazon ā¢^
Of items, of reviewers, of members, of businessesItems rated final āitem ratingā aggregate average of all ratingsItems rated, final item rating aggregate average of all ratingsReviews include text and ratings
Reviews can also be rated and graduates people to āTop 1000ā
reviewer etc.
F^
it^
P^
l^
I fl
ki^
f^
i^
i^
f^
it^
li t
Favorite People. Influence ranking of reviews in favorites list.
-^
IncentivesNone from AmazonPublishers could incent reviewersPublishers could incent reviewers
-^
NegativesBallot stuffing, badmouthing by top reviewersTop reviewer may not be an individual (has to have read more books thanTop reviewer may not be an individual (has to have read more books than
everyone else) Entering the elite circle triggers negative feedbackRatings are cookie-based so can game the system by working around
thatthat
Online Implementations: DiscussionSSpace ā¢^
Slashdot.org
A t
ti^
d^
t^
l^
ti
ā^
Automatic moderator selection
-^
2 layered moderation scheme: M1 for moderating articles, M2 formoderating moderators
-^
The system regularly picks moderators,gives them points to moderate
t^
P^
iti^
/^
ti^
d^
ti^
t^
t^
i fl
th
comments. Positive/negative moderations to comments influence thecomments and the author positively/negatively.
-^
Users have Karma attached to them, karma increases as usersācomments are positively moderated, decreases as they are negativelymoderatedmoderated.
-^
Comments by users with high karma start at a score of 2, low Karmastarts at 0 or -1.
-^
Points given to moderators when they are selected is high or lowd^
di^
th i
k
l^
l
depending on their karma levels.
-^
To address unfair moderations, Slashdot has layer 2 moderators orM2.
-^
Any user can metamoderate several time per day. They will be askedto metamoderate on randomly selected postings. This moderationaffects the Karma of M1 moderators (which in turn impacts their futureability to be moderators)
AdvogatoAdvogato ā¢^
A community of open-source programmersU
t^
t^
h^
t^
i^
b^
d
-^
Uses a trust scheme to manage peer review process basedon PageRank style algorithm (based on a Flow model)Models a flow network (members as nodes and referrals as
edges).
Members refer each other as Apprentice, Journeyer,
Master.
A separate flow graph is generate for each levelA member reachable by the highest level flow graph has
that ratingthat rating
The Reputation MarketThe
Reputation Market
ā¢^
āthelandsellerā case study (Brown, Morgan 2006)āRiddle for a penny! No shipping ā Positive Feedbackā for a penny
- ok: selling a jokesuspicious: title spam āfeedbackā- suspicious: title spam
feedback
- suspicious: total price < cost of listing212 jokes sold (to 172 buyers) at a loss of $87.42At feedback 598 (100%) the seller actually selling land inTexas
Need for Negative Reputation andC
l i
t
Complaints ā¢^
Lack of complaints make reputationimplementations weaker (Resnick 2002)
ā¢^
Lack of penalizing or reducing reputationmechanisms helps create market for tradingrecommendations.(Clausen 2004)SearchKing is a matchmaker of PageRanks (those who have itSearchKing is a matchmaker of PageRanks (those who have it
with those who want it)
Multiple Identities: Sybil Attack onR
t ti
Reputation ā¢^
Sybil Attack: Single person voting many times(Douceur 2002) with multiple identities
ā¢^
So, whatās the cost of an attack on a reputationsystem?
Costs and PayoffsCosts
and Payoffs
ā¢^
For lower pageranks the estimate is tens ofdollars and for high over 100Kā¢
This compares to what SearchKing charges for PageRank
-^
Attacker could buy unmaintained/stale sites for cheap
-^
Attacker could buy unmaintained/stale sites for cheap
-^
Other strategies could be to take over high pagerank sites
ā¢^
High cost of acquiring sites to rip people off maynot make sense. However, once acquired site
q
could scam people with the lack of mechanism forcomplaint
