Download Fuzzy Inference Models for Trust and Reputation Computation in Peer-to-Peer Systems and more Slides Fundamentals of E-Commerce in PDF only on Docsity!
Fuzzy ModelsFuzzy
Models
•^
Use fuzzy inferences to handle uncertainties,f^
i^
d i
l
fuzziness, and incompleteness.Based on the idea that in a P2P transaction
-^
Based on the idea that in a P2P transactionsystem evaluation and dissemination of trust can’tbe effectively done and actors rely on collection of
y^
y
other’s opinions. Global reputation computation istime consuming
-^
2 Major inference stepsLocal Trust InferenceLocal Trust InferenceGlobal Reputation Computation
Trust and Reputation InferenceTrust
and Reputation Inference
•^
Buyer’s local trust score= f(payment method, payment time)
-^
Seller’s local trust score= g(shipping time, goods quality)
-^
Global Reputation weight= h(peer’s trust score transaction a/m transaction date)= h(peer s trust score, transaction a/m, transaction date)Where f, g, h are fuzzy inference functions
Reputation CalculationReputation
Calculation
-^
R
=i ∑
jεS
(w
/∑j
jεS
(w
))tj
ji
=^
∑
jεS
(w
t^ j ji
)/ ∑
jεS
(w
)j
Where R is the reputation score for the Peer i t
is the trust
Where R
is the reputation score for the Peer i, ti
is the trustji^
score of peer i by peer j and w
is the aggregation weight ofj^
t^ ji The global reputation computation is an iterative process and
converges over multiple iterations as a stable reputationscore for peer i
p
Overlay ComputationOverlay
Computation
•^
DHT (Distributed Hash table) algorithm (YideuM i
l 2008)
Mei et al 2008)Each peer maintains 2 tables: a transaction record table and the
peers’ trust scores.The transaction record information is used for computing
weights
To make the algorithm scalable an aggregation threshold is
i^
i^
d^
d^
h^
i h
ib
i^
b l
maintained and peers whose weight contributions are belowthis threshold are not queried for trust scores.
PeerTrust
Liong
,^
Xiu
PeerTrust
Liong
,^
Xiu 2004)
•^
Trust score of a peer is computed as the averageof the scores weighted by the feedback of thepeers
-^
Scores based on 5 factors – peer record,credibility, transaction context, communitycontext and scopecontext, and scope
SmallTrust
(Sakurai Lab, Kyushu
univ
SmallTrust
(Sakurai
Lab, Kyushu univ)
•^
Based on Small World phenomena2 actors in the network are connected by a short path of
acquaintance actors
Static Web (
PageRank
Static
Web (PageRank)
-^
Let P be a set of hyperlinked web pages and let u and vd^
t^
b^
i^
P L t N ( ) d
t^
th
t^
f^
b
denote web pages in P. Let N
-^ (u) denote the set of web
pages pointing to u and N
+^ (v) set of web pages that v points
to. Let be some vector over P that gives an initial rank.
-^
Then the pageRank of a page u is given by:
R(u) = c E(u) + c
∑
vεN-(u)
(R(v)/| N
+^ (v))|
Where c is chosen such that
∑
R(u) = 1
Where c is chosen such that
∑
uεP
R(u) = 1
-^
PageRank applies transitivity of trust to the extreme as trustscores flow through long chains of links.
-^
Personalized PageRank: Vote pages based upon queries:Assigning initial votes based upon the topic of the query(Haveliwala, 2002)(^
)
