Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Recommender systems, specifically collaborative filtering and reputation systems. Collaborative filtering uses the taste of actors to recommend items, while reputation systems provide a common judging mechanism for actors. The document also covers the implementation and combination of these systems, as well as trust models and belief systems.
Typology: Slides
1 / 10
This page cannot be seen from the preview
Don't miss anything!
neighbors in the recommendation space. This information can be used to recommend items that one actorThis information can be used to recommend items that one actor
likes to that actor’s neighbors. Items may be replaced by actors
-^
-^
-^
-^
Centralized systemCentral authority uses a centralized reputation computation
engine E.g. eBay, Amazon, Slashdot,…
-^
Distributed systemP2P system The purpose of reputation system isP2P system. The purpose of reputation system is
Phase 1 (Search phase): to identify which servents (server-
clients) are most reliable at offering the best qualityresources. This may be centralized (Napster)
y^
p^
Phase 2 (Download phase): to identify which servent
provides the most reliable info E.g. KaZaa(Skype), Napster, Gnutella, Freenet,…
g^
yp ),
p^
-^
AccumulativeeBay’s feedback system
Total Positives – Negatives = Feedback scoreTotal Postives/Total
= Feedback percentage
Total Postives/Total
= Feedback percentage
Simple and transparent but gameableEnhanced: weighted schemes based on rater
trustworthiness/reputation, rating age, distance between ratingand current score etc.
Beta-PDF beta(p|
α
,β
) can be expressed using a
function as:
1
β^
1
beta(p|
α
,β
(α
+β
α)
(β
)))p
α^ -1(1-p)
β^
With the restriction that p != 0 if
α
< 1 and p! = 1 if
β
Expectation value of beta distribution is given by
p^
g^
y
E(p) =
α
α^
β)
Reputation can be defined as a function of E(p)The PDF expresses uncertain probability that future interactionsThe PDF expresses uncertain probability that future interactions
will be +ve. Example: Assume a priori distribution of
α
β
After observing some r positive and s negative outcomes, the
posteriori distribution is
α
= r+1,
β
= s+
given r=7, s=1, E(p)=8/10=0.8 meaning that relative frequency of
positive outcome in the future is most likely to be 0.
-^
-^
-^
-^
-^
-^
