Reputation Assessment Framework for Trust Establishment among Web Services Zaki Malik Athman Bouguettaya HungYuan Chung YenCheng Lu Outline Introduction RATEWeb Model Reputation Assessment Techniques ID: 251801
Download Presentation The PPT/PDF document "RATEWeb" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
RATEWeb: Reputation Assessment Framework for Trust Establishment among Web ServicesZaki Malik, Athman Bouguettaya
Hung-Yuan
Chung
Yen-Cheng LuSlide2
OutlineIntroductionRATEWeb ModelReputation Assessment TechniquesExperimentsConclusionSlide3
IntroductionTrust in service-oriented environmentThe web has started a steady evolution to become a “vibrant” environment where applications can be
automatically invoked
by other web clients.
B2C
and
B2B
Business might outsource some of the functionality to other business
We expect enterprises are
no longer a monolithic
organization,
but a coupling of smaller Web-applications
Web services need to determine
which other services can provide the required functionality, before they interact with them. Slide4
Introduction (cont.)There are many web services having the same functionality. They need to compete with
each other.
A mechanism
for the quality access service.
Web services are autonomous, priori unknown, and highly volatile (low
reliability)
Reliable reputation systems increase user’s trust on the Web.
eBay’s feedback Forum, deterring dishonest behavior, and stimulating eBay’s growth.Slide5
RATEWeb (Reputation Assessment for Trust Establishment among Web Services)It provides a comprehensive solution for assessing the reputation of service providers in a reliable, decentralized manner.Different ratings are aggregated to derive a service provider’s reputation.
It takes into account the presence of
malicious raters
that may exhibit oscillating honest and dishonest behaviors.Slide6
Model Entities Web servicesService Providers: a) one provider can provide one or more services
b) a service is provided by a
single
service provider
c) outsource
Services registries
: a collection of descriptions of Web services
Service consumers
(a.k.a. client):
i
nvokes a Web service
A human user uses a Web service
proxy
. The human user only communicates his/her needs to the service proxy, and all decisions are all taken by the service proxy. (everything is automated)Slide7
Scenario: Car Brokerage ApplicationA company deploys a car broker Web service (CB)
CB is registered with service registries (Then consumer can obtain details through the registry)
CB may outsource from other web services.
e.g., car dealer, lemon check, financing, credit history, insurance
S
ervice providers may also act as consumers.
A consumer access a CB service to buy a car. Then a series of invocations would need to take place.
The selection of a service by CB at each invocation step can be done in two ways:
with or without
reputation systemSlide8Slide9Slide10
ComparisonNo guarantees about the delivery of the required functionality could be made before the actual interaction.Scenario 1: (one monopoly)
From the consumer’ respect, the scenario described is far from optimal.
Scenario 2: (competition among CBs)
The providers can use service’s reputation when composing their CBs
.
CB can reduce the risk of its own reputation getting tarnished.
Consumers can select the
best CB
based on the
different
CB’s individual reputation.Slide11
Extension: Community Community: a container that clumps together Web services related to a specific area of interestAll Web service that belongs to a given community share the same area of interest.
Responsibilities:
Set reputation
threshold
.
Set
rules
when a member’s reputation goes below the threshold.
Define reputation requirements for new members. Slide12
DefinitionCommunity ci := (Identifier
i
,
Category
i
, Generic-
operation
i
,
Members
i
)
Identifier
i
: contains name and features of c
i
Category
i
: contains areas of interests
G-
operations
i
: summarizes the major functions needed by community members
Member
i
: a list of members. Members will support one or several of
c
i
‘s
generic operation Slide13
Model InteractionsService providers can register their web services with communities.The consumer can access service registries to get the details of a communities and providers.Communities search their directories for the list of providers that have registered their operations.
Communities also contain a list of consumers that had interacted with each members in the past.
The consumer then selects the best provider form the list.
The community only act as a directory of raters
not as a centralized repository of rating
ratings are
keep local
with the ratersSlide14Slide15
Reputation Assessment Parameters reflecting the Quality of Web Services:
Provider-promised
Consumer-expected
Service-delivered
Quality parameter
is the
kth
quality parameter
When a service requester
invokes the service
, each quality parameter
in
gets assigned a
delivered quality value
Slide16
Web Service Reputation : The set of service consumers
: Personal evaluation,
represents only consumer
’s perception
of
the
provider
’s reputation
: Aggregation function
Slide17
Reputation Evaluation MetricsRater CredibilityMajority RatingPast Rating HistoryPersonal Experience for Credibility EvaluationPersonal PreferencePersonal Experience for Reputation AssessmentTemporal SensitivitySlide18
Reputation Evaluation MetricsRater CredibilityMajority RatingPast Rating HistoryPersonal Experience for Credibility EvaluationPersonal PreferencePersonal Experience for Reputation AssessmentTemporal SensitivitySlide19
Rater CredibilityIn order to cater for such bad-mouthing or collusion possibilities, the system should weigh highly credible raters than low credible raters
How to get the
s?
Slide20
Rater CredibilityIdea 1: “if the reported rating agrees with the majority opinion, the rater’s credibility is increased, and decreased otherwise”Majority opinion:
By K-means clusteringSlide21
Rater CredibilityThe change in credibility due to majority rating, denoted by is defined as:
where
is the standard deviation in all the reported ratings and
is the reported rating (of each rater),
note that the k is different from the clustering
In short: deduce more credibility if your opinion is differentSlide22
Rater CredibilityIdea 2: difference with the opinions in a time periodNote that k has different meanings in the 2 eqs
.
k
: valid time lag
t: current timestampSlide23
Rater Credibility
Based on
, the authors suggest several ways to estimate the credibility
General form:
is the credibility adjustment normalizing factor
is credibility change due to
: “pessimism factor” –
low-> optimistic
high->pessimistic
: “pessimism factor” –
low-> pessimistic
high-> optimistic
Slide24
Rater CredibilityUsefulness factor – “The usefulness of a service is required to calculate a service rater’s “propensity to default,” i.e., the service rater’s tendency to provide false/incorrect ratings.”
where
Ui
is the
submission where the rater was termed “
useful”
and
Vx
denotes the
total number of ratings
submissions by
that service
.Slide25
Personalized Preferences
:
the
rating assigned to attribute
by the
service rater
for service provider
in transaction
,
:
the
total number of attributes
:
the preference
of the service consumer for attribute
Slide26
Temporal SensitivityReputation fader – fade out the out-dated ratingsE.g., where
is the total number of
past
transactions
over
which the
reputation is to be evaluated
Slide27
First-hand knowledgeFinally,Slide28
Reputation AssessmentSlide29
Experimental Evaluations ()Parameter settingsSlide30
# High credibility >> # Low credibilitySlide31
# High credibility = # Low credibilitySlide32
# High credibility << # Low credibilitySlide33
Low (Optimistic consumer)
Slide34
High (Pessimistic consumer) Slide35
Transaction Success RateSlide36
Reputation ErrorSlide37
Cost Analysis ExperimentsRuntime overhead mainly involves Retrieving required information Assimilate all the gathered informationThe cost is directly influenced by the reputation collection model used.Publish-subscribe modelCommunity broadcast modelCredibility-based modelSlide38
Publish-subscribe modelSlide39
Community broadcast modelSlide40
Credibility-based modelSlide41
Cost Analysis Parameter settings:Slide42
Cost AnalysisSlide43