UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

UNITIV PREDICTINGHUMANBEHAVIOURANDPRIVACYISSUES
Understandingandpredictinghumanbehaviourforsocialcommunities-Userdatamanagement
-Inference and Distribution - Enabling new human experiences - Reality mining - Context -
Awareness - Privacy in online social networks - Trust in online environment - Trust modelsbased
on subjective logic - Trust network analysis - Trust transitivity analysis - Combining trust and
reputation - Trust derivation based on trust comparisons - Attack spectrum and countermeasures.
Understandingandpredictinghumanbehaviorforsocialcommunities
With the rapid advance in technology, it is becoming increasingly feasible for people to take
advantageofthedevicesand services in the surrounding environment to remain “connected”and
continuously enjoy the activity they are engaged in, be it sports, entertainment, or work.
Ubiquitous computing environment will allow everyone permanent access to the Internet
anytime, anywhere and anyhow.
QoE (Quality of Experience) is a consequence of a user‟s internal state (e.g., predispositions,
expectations, needs, motivation, mood), the characteristics of the designed system (e.g.,usability,
functionality, relevance) and the context (or the environment) within which the interaction
occurs (e.g., social setting, meaningfulness of the activity).

UserDataManagement,InferenceandDistribution
 User-oriented creation/execution environments lack on the capability to adapt to the
heterogeneity of devices, technologies and the specificity of each individual user.
 For user flexibility and personalization requires user profile management systems which
include limited information about user preferences and contexts.
 In order to apply user information across a range of services and devices, there is a need
for standardization of user related data and the architecture that enables their
interoperability.
 These efforts have been taken by European Telecommunications Standards Institute
(ETSI), the Third Generation Partnership Project (3GPP), Open Mobile Alliance (OMA)
 Considering data requirements from a wide range of facilities the concept of Common
Profile Storage (CPS) is defined by 3GPP.
 UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
o as a framework for streamlining service-independent user data and storing itunder
a single logical structure in order to avoid duplications and data inconsistency
 Logicallycentralized data storage can be mapped to physicallydistributed configurations
and should allow data to be accessed in a standard format
 Data storage can be grouped into three main classes: the syntactic, semantic
andmodelingapproaches whichenable interoperabilityofuserprofile data management fora
Future Internet.
 To improve the degree of services personalization it is important to generate new
information from the existing one.
 In this sense, social networks, user modeling and reality mining techniques can be
empowered to study patterns and predict future behaviors.
 The basic motivation is the demand to exploit knowledge from various amounts of data
collected, pertaining to social behavior of users in online environments.
 To handle complex situations, the concept of decomposition is applied to the situation
into a hierarchy of sub-situations.
 These sub-situations canbe handled autonomouslywithrespect to sensing andreasoning.
The handling of complex situations can be simplified by decomposition.
 Anothersimilarperspectiveiscalledlayeredreasoning,where
o the first stage involves feature extractionand grouping (i.e., resulting in low-level
context),
o thesecondevent,stateandactivityrecognition(i.e.,originatingmid-level context),
while
o thelaststageisdedicatedtopredictionandinferenceofnew knowledge
 ResearchinSocialnetworkusuallyfocuseson
o quantifyingorqualifyingtherelationshipbetweenpeers,
o wherealgorithmssuchascentralityandprestigecanbeusedtocalculatethe proximity,
influence or importance of a node in a network,
o whileclusteringandclassificationcanbeappliedtosimilaritycomputation,respectively
.
 UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
Combining allofpre-enunciated conceptswithontologies and semantic technologies, wepresent a
generic framework for managing user related data, will pave the way to understanding and
predicting future human behavior within social communities.

EnablingNewHumanExperiences
Itisimportanttounderstandwhat arethetechnologiesbehind userdatamanagement,howto link them
and what can they achieve when combined in synergy.
The Technologies
a. SocialNetworks
 Humansinallculturesatalltimesformcomplexsocial networks
 Social network - means ongoing relations among people that matter to those engaged in
the group, either for specific reasons or for more general expressions of mutual
agreement.
 Social networks among individuals who may not be related can be validated and
maintained byagreement onobjectives, socialvalues, or even bychoiceofentertainment.
o involve reciprocal responsibilities and roles that may be selfless or self-interest
based.
 Social networks are trusted because of sharedexperiences and the perception of
sharedvalues and shared needs
 Behaviorofindividualsinonlinenetworkscanbeslightlydifferentfromthesameindividuals
interacting in a more traditional social network (reality).
 Itgivesusinvaluableapproachesonthepeoplewearecommunicatingwith,which groups are
we engaged, which are our preferences, etc.

b. RealityMining
 To overcome the differences between online and “offline” networks, reality mining
techniques can be empowered to approximate both worlds, proving awareness about
people actual behavior.
 It is the collection and analysis of machine sensed environmental data pertaining to
human social behavior.
 Ittypicallyanalyzessensordatafrommobiles,videocameras,satellites,etc
 UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

 Predictivepatternssuchas„honestsignals‟providemajorfactorsinhumandecision making.
 Realityminingenables„bigpicture”ofspecificsocialcontextsbyaggregatingand averaging
the collected data
 Itallowsdata/eventscorrelationandconsequentlyfutureoccurrencesextrapolation.

c. Context-Awareness
Byassessingandanalyzingvisionsandpredictionsoncomputing,devices, infrastructuresand human
interaction, it becomes clear that:
a. contextisavailable,meaningful,andcarriesrichinformationinsuchenvironments,
b. thatusers‟expectationsanduserexperience isdirectlyrelatedtocontext,
c. acquiring,representing,providing,andusingcontext becomesacrucialenabling technology for
the vision of disappearing computers in everyday environments.

4.3.2ArchitecturalFramework and Methodology

Toenablehumanbehaviorunderstandingandprediction,thereareseveralindependentbut
complementary steps that can be grouped into three different categories:
 DataManagement
 NewKnowledgeGenerationand
 ServiceExposureand Control.
a. Datamanagement
 This activityusuallystarts withdataacquisition.
 Involvesgatheringofinformationfromdifferentinformationsystems.
 Figuredepictstheserelationshipsaswellasthesequenceofactivitiesinvolved.
 UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

 Figurebelowexemplifiesthe typeofinformationthat canbestoredintheHumanData

Repository
 UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

 Data is not usually captured without errors. Therefore it is necessary to preprocess
itinadvance before mining.
 Otherwiseitwouldnotbepossibleto correlateinformationcorrectly.
 Oncethis isdone,datais minedbyusingtwodifferentapproaches:
 knowstatisticalalgorithmsto help patternrecognitionandconsequent algorithmic
modeling,
 the opposite approach, where specific algorithms are designed to identifypatterns
in the data (this requires previous modeling).
Combiningboth,allowsustoaddressthespecificsofourapplications,andatthesametime, automatically
detect new relevant correlations that might occur after a few iterations.

b. KnowledgeGeneration
 Newinformationinferenceisbasedonuserrelateddata(calledascontext)
 Threedifferent categories:
o Real-time
o Historicaldata
o Reasonedcontext
 In Fig. below, there are several layers of abstraction in a context-aware system and any
context-aware middleware or architecture must therefore be capable of building
representations and models of these abstractions.
 However, thesehigh-levelabstractionscanonlybe made from lower levelcontext, which
requires some form of context management function (performed by a Context Broker).
 Inourcase,thisisperformed at theHuman DataRepository.
 The main context management features are context acquisition, context aggregation &
fusion, context dissemination, discovery and lookup.
 In order to manipulate context information, it must be represented in some form that is
compatible with the models that will be used in the reasoning and situation recognition
processes.
 These models could be object oriented, ontological, rule based, logic based, based on
semantic graphs or fuzzy logic sets.
 Reasoning mechanisms allow high-level context to be deduced or situations to be
recognized that is output of one process can be used as an input to another.
 UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
 Reasoning isalsousedtochecktheconsistencyofcontextandcontextmodels.
 It is very important to stress that the prediction does not necessarily anticipates the user
wishes or desires, but a possible future that could be interesting for the user.

c. ServiceExposureand Control
 Thethirdlayerisdividedintotwomaincapabilities.
 The first is user-centric and relates to the ability of the user to stay in control of the
whole scenario, enabling it to specify when, what, why, who, where and how the data is
or can be accessed.
 Through the Human Enabler, users are able to influence the way their behavior is
predicted, by controlling how there are being profiled (automatic, off, manually
personalized).
 This is essentialforestablishing and managingtrust and forsafeguardingprivacy, aswell as
for designing and implementing business security models and policies.
 The second set of features is associated with the capacity of exposing thisinformation
(both raw data and inferred one) to third party service providers (such as advertising
agencies), through well defined web service interfaces.
 Besides exposing user related information, the human enabler allows data to be
subscribed, syndicated or updated on request.

4.3.3.Innovations
Theanalysisofthefirstresultsindicatedthefollowingkeyfindings:
 Itispossibletoinferuserbehaviorbasedonuserpreferences,socialnetworksand context-aware
systems, with the help of reality/data mining techniques.
 Proximity and Similarity are greatweightindicators forinferringinfluence andcan be
computed or calculated analytically.
 Bothonlineandofflinesocialnetworkshaveinfluenceoveraperson‟s behavior.
 UserperceivedQoEisimprovedasthemethodologydeliverspersonalization,
contextualization, interactivity, adaptation and privacy.
 Usersarewillingtoparticipateintheirownprofilingexperienceandtheresultsare positive.
 UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
Applying these techniques into different fields of computer social sciences may have significant
applicability in different parts of the value chain.
Examples:
 Inferandsuggestmissinginformationinusersprofileaccordingtohis/herpeers contextual
information.
 Understand howaspecific usercanbe influenced byanotheruserorcommunity and vice
versa.
 Understandhowsimilartwousersare, eveniftheydonothavefriends incommon.

Privacyin OnlineSocialNetworks
 There is adramaticgrowth in number and popularityofonline social networks.Thereare
many networks available with more than 100 million registered users such as Facebook,
MySpace, QZone, Windows Live Spaces etc.
 People may connect, discover and share by using these online social networks. The
exponential growth of online communities in the area of social networks attracts the
attention of the researchers about the importance of managing trust in onlineenvironment.
 Users of the online social networks may share their experiences and opinions within the
networks about an item which may be a product or service.
 Collaborativefilteringsystemisthe mostpopularmethodinrecommender system.
o Thetaskisto predicttheutilityofitemsto aparticular userbasedonadatabaseof user
rates from a sample or population of other users.
 Because of the different taste of different people, they rate differently according to their
subjective taste.
 If two people rate a set of items similarly, they share similar tastes. In the recommender
system, this information is used to recommend items that one participant likes, to other
persons in the same cluster.
 Performs poor when there is insufficient previous common rating available betweenusers;
known as cold start problem
 To overcome the cold start problem trust based approach to recommendation has
emerged.
 UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
 This approach assumes a trust network among users and makes recommendations based
on the ratings of the users that are directly or indirectly trusted by the target user.
 Trustcouldbeusedassupplementaryorreplacementofcollaborativefilteringsystem
 Trust and reputation systems can be used in order to assist users in predicting and
selecting the best quality services
 Binomial Bayesian reputation systems normally take ratings expressed in a discrete
binary form as either
o positive(e.g.good)or
o negative(e.g.bad).
 Multinomial Bayesian reputation systems allow the possibility of providing ratings with
discrete graded levels such as e.g. mediocre – bad –average – good – excellent
 Trust models based on subjective logic are directly compatible with Bayesian reputation
systems because a bi-jectivemapping exists between their respective trust and reputation
representations.
 This provides a powerful basis for combining trust and reputation systems for assessing
the quality of online services.
 Trust systems can be used to derive local and subjective measures of trust, meaning that
different agents can derive different trust in the same entity.
 Reputation systems compute scores based on direct input from members in the
community which is not based on transitivity
 Bayesian reputation systems are directly compatible with trust systems based on
subjective logic, theycanbe seamlessly integrated. Thisprovidesa powerfuland flexible
basis for online trust and reputation management.

OnlineSocialNetworks
 A social network is a map of the relevant ties between the individuals, organizations,
nations etc. being studied.
 With the evolution of digital age, Internetprovides a greater scope of implementing social
networks online. Online social networks have broader and easier coverage of members
worldwide to share information and resources.
 The first online social networks were called UseNet Newsgroups. designed and built by
Duke University graduate students Tom Truscott and Jim Ellis in 1979.
 UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
 Facebook is the largest and most popular online social network at this moment
(www.insidefacebook.com).
 It had350 millionMonthlyActiveUsers (MAU) atthe beginningofJanuary2010. But it has
been growing too fast around the world since then.
 As on 10 February 2010, roughly 23 million more people are using Facebook compared
to 30 days ago, many in countries with big populations around the world. This is an
interesting shift from much of Facebook‟s international growth to date.
 Once Facebook began offering the service in multiple languages it started blowing up in
many countries like Canada, Iceland, Norway, South Africa, Chile, etc.
 The United States is at the top with more than five million new users; it also continues to
be the single largest country on Facebook, with 108 million MAU

TableaToptenmostlyvisitedsocialnetworksinJan‟09–basedonMAU

TrustinOnline Environment
 Trusthasbecomeimportanttopicofresearchinmanyfieldsincludingsociology, psychology,
philosophy, economics, business, law and IT.
 Trustisacomplexwordwithmultiple dimensions.
 Though dozens of proposed definitions are available in the literature, a complete formal
unambiguous definition of trust is rare.
 Trust isusedasawordorconceptwith norealdefinition.
 Trust is such a concept that crosses disciplines and also domains. The focus ofdefinition
differs on the basis of the goal and the scope of the projects.
 Two forms
 UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
o reliabilitytrustorevaluation trust
o decisiontrust
 Evaluation trust canbe interpretedasthereliabilityofsomethingor somebody. It canbe
defined as the subjective probability by which an individual, A, expects that another
individual, B, performs a given action on which its welfare depends.
 The decision trust captures broader concept of trust. It can be defined as the extent to
which one party is willing to depend on something or somebody in a given situation with
a feeling of relative security, even though negative consequences are possible.

TrustModelsBasedonSubjectiveLogic
 Subjective logic is a type ofprobabilistic logic that explicitlytakes uncertainty and belief
ownership into account.
 Argumentsinsubjectivelogicaresubjectiveopinionsaboutstatesinastatespace.
 A binomial opinion applies to a single proposition, and can be represented as a Beta
distribution. A multinomial opinion applies to a collection of propositions, and can be
represented as a Dirichlet distribution.
 Subjective logic defines a trust metric called opinion denoted by which
expresses the relying party A‟s belief over a state space X.
o Here represents belief masses over the states of X, and u represent uncertainty
mass where, u €
o The vector

o represents the base rates over X, and is used for computing the probability
expectation value of a state x.

 Binomial opinions are expressed as where d denotes disbelief in x.

When the statement x for example says “David is honest and reliable”, then the opinion
can be interpreted as reliability trust in David.
 Let us assume that Alice needs to get her car serviced, and that she asks Bob to
recommend a good carmechanic. WhenBobrecommendsDavid, Alice would like to get
 UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
asecondopinion,sosheasksClaireforheropinionaboutDavid.Thissituationis illustrated in Fig.
4.6 a

Fig.4.6aDerivingtrustfromparalleltransitive chains

Whentrustandreferralsareexpressedassubjectiveopinions, eachtransitivetrustpath
Alice →Bob→ David→ and Alice → Claire → David can be computed with the transitivity
operator, where the idea is that the referrals from Bob and Claire are discounted as a function
Alice‟s trust in Bob and Claire respectively. Finally the two paths can be combined using the
cumulative or averaging fusion operator. These operators form part of Subjective Logic and
semantic constraints must be satisfied in order for the transitive trust derivation to bemeaningful.

 Thismodelisthusbothbelief-basedand Bayesian.
 A trust relationship between A and B is denoted as [A:B]. The transitivity of two arcs is
denoted as“:” and the fusion oftwo parallelpaths is denoted as“◊”. The trust network of
Fig. 4.6 a can then be expressed as:
[A,D] =([A,B]:[B,D])◊([A,C]:[C,D])
 Thecorrespondingtransitivityoperatorforopinionsdenotedas“ ”andthecorresponding
fusion operator as “ ”. The mathematical expression for combining the opinions about
the trust relationships of Fig. 4.6 a is then:
 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

 Arbitrarily complex trust networks can be analysed with TNA-SL which consists of a
network exploration method combined with trust analysis based on subjective logic.
 The method is based on simplifying complex trust networks into adirected seriesparallel
graph (DSPG) before applying subjective logic calculus.

Trustnetworkanalysis
 Trust networks consist oftransitive trust relationships between people, organizations and
software agents connected through a medium for communication and interaction.
 Trust network analysis using subjective logic (TNA-SL) takes directed trust edges
betweenpairs as input, and can be used to derive a leveloftrust betweenarbitraryparties
that are interconnected through the network.
o Incaseofno explicit trust pathsbetweentwo partiesexist;subjective logicallows a
level of trust to be derived through the default vacuous opinions.
o TNA-SL issuitableformanytypesoftrustnetworks.
o Limitation:complextrustnetworksmustbesimplifiedtoseries-parallel
networksinorderforTNA-SLtoproduceconsistentresults.
o Thesimplification consistedof gradually removingtheleastcertaintrustpaths until
the whole network can be represented in a series-parallel form.
o Asthisprocessremovesinformationitisintuitivelysub-optimal.
OperatorsforDerivingTrust
 Subjectivelogicisabeliefcalculusspecificallydevelopedformodelingtrust relationships.
o Beliefsarerepresentedonbinarystatespaces,whereeachofthetwopossible states can
consist of sub-states.
o Belieffunctionsonbinarystatespacesarecalledsubjectiveopinions
 Expressedintheformofanorderedtuple

u ∈ [0, 1] and b+d+u = 1

where b,d,andurepresent belief,disbeliefanduncertaintyrespectivelywhere b, d,

 Thebaserateparametera∈[0,1]representsthebaserateprobabilityin theabsenceof evidence,

and is used for computing an opinion‟s probability expectation value

13 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

 Asubjectiveopinionis interpretedasanagentA‟sbeliefinthetruthofstatementx
 A‟sopinionabout xisdenotedas

 Subjective logic defines a rich set of operators for combining subjective opinions in
various ways. Some operators represent generalizations of binary logic and probability
calculus, whereas others are unique to belief calculus because they depend on belief
ownership.
Transitivity is used to compute trust along a chain of trust edges. Assume two agents A and B
whereAhasreferraltrustinB,denotedby ,forthepurposeofjudgingthefunctionalor

referraltrustworthiness ofC.
In addition B has functional or referral trust in C, denoted by . Agent A can then derive her

trust in C bydiscounting B‟s trust in C with A‟s trust in B, denoted by .

Byusingthesymbol“ ”todesignate thisoperator,we define

CumulativeFusion isequivalent toBayesianupdating instatistics.The cumulative fusionoftwo

possiblyconflicting opinions is an opinion that reflects both opinions in a fair and equal way
Let and be A‟sandB‟strustinC respectively. Theopinion isthencalledthefused trust

between and , Byusing the symbol “ ” to designate this operator, we define

14 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

Theeffectofthecumulativefusionoperatoristoamplifybeliefanddisbeliefandreduce uncertainty

TrustPathDependencyandNetworkSimplification
 Transitivetrustnetworkscaninvolvemanyprincipals
 CapitallettersA,B,CandDwillbeusedtodenoteprincipals.
 Asingletrustrelationshipcanbeexpressedasadirectededgebetween twonodesthat represent
the trust source and the trust target of that edge.
 For example the edge [A, B] means that Atrusts B. The symbol “:” is used to denote the
transitive connection of two consecutive trust edges to form a transitive trust path.
 The trust relationships between four principals A, B, C and D connected serially can be
expressed as:
([A,D])=([A,B]:[B,C]:[C,D])
 Wewillusethesymbol“◊”todenotethegraphconnector.The“◊”symbolvisually resembles a
simple graph of two parallel paths between a pair of agents.
 In shortnotation,A‟s combination of thetwoparallel trustpathsfrom hertoDisthen
expressed as:
([A,D])=(([A,B]: [B,D])◊([A,C]:[C,D]))
 Trustnetworkscanhavedependentpaths.Thisisillustratedontheleft-handsideofFig.
4.7.a.Theexpressionforthegraphontheleft-handsideofFig. 4.7.awouldbe: ([A,D])
= (([A,B] : [B,D]) ◊ ([A,C] : [C,D]) ◊ ([A,B] : [B,C] : [C,D]))
 Trustnetworkanalysiswithsubjectivelogicmayproduceinconsistentresultswhen applied
directly to non-canonical expressions.

Fig.4.7.aNetworksimplificationbyremovingweakestpath
It is therefore desirable to express graphs in a form where an arc only appears once. A canonical
expression can be defined as an expression of a trust graph in structured notation where every
edge only appears once.

15 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

Assuming that thepath([A,B]:[B,C]:[C,D])isthe weakest path inthegraphonthe left-hand side of
Fig. 4.7.a, network simplification of the dependent graph would be to remove the edge [B,C]
from the graph, as illustrated on the right-hand side of Fig. 4.7.a.

TrustTransitivityAnalysis
Assume two agents Aand B where Atrusts B, and B believes that proposition x is true. Then by
transitivity, agent Awillalso believe that propositionx is true. This assumes that Brecommends x
to A. In our approach, trust and belief are formally expressed as opinions. The transitivelinking
of these two opinions consists of discounting B‟s opinion about x by A‟s opinion about B, in
order to derive A‟s opinion about x. This principle is illustrated in Fig. 4.8.a below

Fig.4.8.aPrincipleoftrusttransitivity solid
arrows - initial direct trust
dottedarrow- derivedindirecttrust

UncertaintyFavoringTrustTransitivity
A‟sdisbeliefintherecommendingagentBmeansthatAthinksthatBignoresthetruth valueof
x. AsaresultAalsoignoresthetruthvalueofx.
UncertaintyFavoring Discounting
LetAand Bbetwo agents
whereA‟sopinionaboutB‟srecommendationsisexpressedas

letxbeapropositionwhereB‟sopinionaboutxisrecommendedtoAwiththeopinion

Let

16 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

-theuncertaintyfavoringdiscountedopinionofA. By
using the symbol

This operator is associative but not commutative. This means that the combination of opinions
can start in either end of the path, and that the order in which opinions are combined is
significant.
Figurebelow4.8.billustratesanexampleofapplyingthediscountingoperatorforindependent
opinions, where
discounts toproduce

Fig.4.8.bExampleofapplyingthediscountingoperatorforindependentopinions

OppositeBeliefFavoring
A‟s disbelief in the recommending agent B. A not only disbelieves in x to the degree that B
recommends belief, but she also believes in x to the degree that B recommends disbelief in x,
because the combination of two disbeliefs results in belief in this case.
OppositeBeliefFavoringDiscounting
LetAandBbetwoagentswhereA‟sopinionaboutB‟srecommendationsisexpressed as

17 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

letxbeapropositionwhereB‟sopinionaboutxisrecommendedtoAwiththeopinion

Let

-oppositebelieffavoringdiscountedrecommendationfromB toA By
using the symbol

This operator models the principle that “your enemy’s enemy is your friend”. It isdoubtful
whether it is meaningful to model more than two arcs in a transitive path with this principle. In
other words, it is doubtful whether the enemy ofyour enemy‟s enemy necessarily is your enemy
too.
BaseRateSensitive Transitivity
Imagineastrangercomingtoatownwhichis knownfor itscitizensbeing honest. Thestranger is
looking for a car mechanic, and asks the first person he meets to direct him to a good car
mechanic. The stranger receives the reply that there are two car mechanics in town, David and
Eric, where David is cheap but does not always do quality work, and Eric might be a bit more
expensive, but he always does a perfect job.
Translated into the formalismofsubjective logic, the stranger has no other info about the person
he asks than the base rate that the citizens in the town are honest. The stranger is thus ignorant,
but the expectation value of a good advice is still very high.
Without taking into account, the result of the definitions above would be that the stranger is

completelyignorantaboutwhichofthemechanicsisthebest.Anintuitiveapproachwouldthen

18 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

betolettheexpectationvalueofthestranger‟strustintherecommenderbethediscounting factor for the
recommended parameters.

BaseRate SensitiveDiscounting
Thebaseratesensitivediscountingofabelief

byabelief

Producesthetransitivebelief

probabilityexpectationvalueE

Asafetyprinciplecouldthereforebeto onlyapplythebaseratesensitivediscountingtothelast transitive

link.

Mass Hysteria
Consider how mass hysteria can be caused by people not being aware of dependence between
opinions. Let‟stake for example;personArecommend anopinionabout aparticular statement x to
agroupofother persons. Without being aware ofthe fact that theopinioncame fromthesame origin,
these persons can recommend their opinions to each other as illustrated in Fig. 4.8.c

19 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

Fig. 4.8.c Theeffectsofunknowndependence

Thearrowsrepresenttrust sothatforexampleB ->Acanbe interpretedassayingthat BtrustsA to

recommend an opinion about statement x. The actual recommendation goes, ofcourse, in the
opposite direction to the arrows in Fig. 4.8.c.
IfGassumestherecommendedopinionstobeindependentandtakestheconsensusbetween them, his
opinion can become abnormallystrong and in fact even stronger than A‟s opinion.
In orderto reducethe size ofthe notation, the transitivity symbol “:” will simply be omitted, and
the cumulative fusion symbol ◊ will simply be written as “,”. Analyzing the whole graph of
dependent paths, as if they were independent, will then produce:

Whenthisprocesscontinues, anenvironmentofselfamplifyingopinions,andtherebymass hysteria, is

created.

Combiningtrust andreputation
TheDirichletReputationSystem

20 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV

CS6010 – SocialNetworkAnalysis– UnitIV

Reputationsystemscollectratingsaboutusersorserviceprovidersfrommembersina community.
Thereputationcentreisthenableto computeand publishreputation scoresabout thoseusersand
services.

Figure4.9.aSimplereputationsystem
Multinomial Bayesian systems are based on computing reputation scores by statistical updating
of Dirichlet Probability Density Functions (PDF)
A posteriori (i.e. the updated) reputation score is computed by combining a priori (i.e. previous)
reputation score with new ratings.
Agentsareallowedto rateothersagentsor services withanylevelfroma set ofpredefinedrating levels
Reputationscoresarenot staticbutwillgraduallychange
Let therebek different discreterating levels.Let therating levelbe indexed byi. Theaggregate
ratings for a particular agent can be expressed as

This vector can be computed recursively and can take factors such as longevity and community
base rate into account.
-aggregateratingofaparticularlevelifor agenty
Beforeanyratingsaboutaparticularagentyhavebeenreceived,itsreputationisdefinedby common base
rate
Ratings about particular agent are collected, the aggregate ratings can be computed
recursivelyand derived scores will change accordingly.
ThevectorSisdefinedby

21 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
ThereputationscoreSisdefinedby

To express reputation score as a single value in some predefined interval. This can be done by
assigning a point value to each rating level L :

A bijective mapping can be defined between multinomial reputation scores and opinions, which
makes it possible to interpretthesetwo mathematicalrepresentations asequivalent. The mapping
can symbolically be expressed as:

Theorem:EquivalenceBetweenOpinionsandReputations
Let w=(b,u,a) be an opinion, and R be a reputation, both over the same state space X so that the
base rate „a‟ also applies to the reputation. Then the following equivalence

Multinomial aggregate ratings can be used to derive binomial trust in the form of an opinion.This
isdone byfirst convertingthe multinomialratingsto binomial ratings according to equation below
and then apply to the above theorem.
Thederived converted binomialrating parameters(r,s)aregivenby:

22 B.ShanmugaSundari, AP/IT, PETEngineering College

 csenotescorner.blogspot.com UNITIV
CS6010 – SocialNetworkAnalysis– UnitIV
Figure4.9.billustratesascenariowhereagentAneedstoderiveameasureoftrustinagentF.

AgentBhasreputationscore (arrow1),andagentAhas trust

RCintheReputationCentre (arrow 2),sothatAcanderiveameasure oftrustinB(arrow 3).

Agent B‟s trust in F (arrow 4) can be recommended to Aso that Acanderive ameasure oftrust in
F (arrow 5). Mathematically this can be expressed as:

The compatibility between Bayesian reputation systems and subjective logic makes this a very
flexible framework for analysing trust in a network consisting of both reputation scores and
private trust values.

23 B.ShanmugaSundari, AP/IT, PETEngineering College