Modeling and Representing Trust Relations in Semantic Web-Driven Social Networks: An Ontological Analysis

Nima dokoohaki

Royal Institute of Technology,Stockholm, June 2007

Agenda

• Trust in Semantic web• Ontologies• Engineering of trust

ontology– Development– Analysis

• Conclusion• Future work• Questions

Semantic web

• An architectural plan to augment the existing web

• Limitation of current web– Lack of Semantics for software

agents

• How:By Using a set of semantic standards

• Goal: Giving the information on the web more meaning

• Main Reason– More information accessibility

and availability on the web

Adopted from “The Semantic Web: A Guide to the Future of XML, Web Services, and Knowledge Management” – Daconta et al., 2003

StructureStructure XML SchemaXML Schema

Security and IdentitySecurity and Identity

Inference EngineInference Engine

DAML+OIL,OWLDAML+OIL,OWL

RDF/RDF SchemaRDF/RDF Schema

Reasoning and ProofReasoning and Proof

Higher SemanticsHigher Semantics

SemanticsSemantics

TrustTrust

Syntax DataSyntax Data XMLXML

Domain NameSpaceDomain NameSpace

Intelligent Domain Applications and ServicesIntelligent Domain Applications and Services

Importance of trust in semantic web• Semantic web is an

environment with certain features like; – Ubiquitous, – Heterogeneous, – Openness.

• In such an environment, trusting other people and their contributions becomes of great importance.

Trust Definition

Tyron Grandison’s definition:

• Trust is ” the firm belief in the competence of an entity to act dependably, securely, and reliably within a specified context”.

• Distrust is defined as the Lack of Trust.

Trust properties• Trust is described to be: According to

Grandison, and Abdul-Rahman et al.

1. Purpose or context-dependant. For instance, Alice trusts Bob as a doctor, but she might not trust Bob as a car mechanic.

2. Quantitative or qualitative (metric). represents the intensity and level of trust. For instance, Alice might trust Bob as a doctor very much, while she only moderately trusts Martin as a doctor.

3. Transitive, or Non-transitive . when Alice trusts Bob and Bob trusts Anna, Alice will trust Anna.

4. Dynamic. Trust is dynamic and it is permanently changing.

5. Non-monotonic. Further observations may elevate or lower the level of trust invested in another entity.

Trust properties contd.• Two distinctions are made;• Distinction 1:

1. trust in performance: trust in an entity to perform an action,• Alice trusting Bob as a doctor

2. trust in recommendation: trust in an entity to recommend other entities to perform that action.• Alice trusting Bob to recommend a

good doctor

• Distinction 2: Considering existence of recommenders1. trust from direct observation of

the trustee 2. trust derived from the

observations by the recommenders

Social Networks• WBSN, Web-based social

networks• Evolved beyond text messages• Ability to specify and state

reliability, trust and belief• Survey done by Jen Golbeck

done in 2005: “115,000,000 members of the social networks discovered across about 18 networking societies .“– Another big bang is quite close!

• Solution: integration and merging of the data distributed across these networks

Semantic Social Networks • SSN, describes the notion of

integrated and merged social networks. [Stephen Downes, 2004]:– “Network resources are

expressed in XML or RDF, such as descriptions of persons (authors, readers, critics)

– References to those descriptions employed in RDF or XML files describing resources.”

• potential solution for merging and integration of information– The Friend-of-a-Friend (FOAF)

Project [Dumbill, 2002]

Foaf: A framework ,a vocabulary and an ontology• FOAF is “a framework for

representing information about people and their social connections”. Golbeck’s definition

• The FOAF Vocabulary [Brickley, Miller, 2004] contains terms for describing and depicting personal information (such as name, surname and email address), membership in groups and social connections

• FOAF is represented as an ontology, using RDF and OWL

Sample Foaf profile<foaf:Person rdf:ID="me">

<foaf:name>Nima Dokoohaki</foaf:name><foaf:title>Mr</foaf:title><foaf:givenname>Nima</foaf:givenname><foaf:family_name>Dokoohaki</foaf:family_name>

<foaf:depiction rdf:resource="http://nimadokoohaki.googlepages.com/nimatbana.JPG"/><foaf:workplaceHomepage

rdf:resource="http://nimadokoohaki.googlepages.com"/><foaf:schoolHomepage

rdf:resource="http://www.imit.kth.se"/><foaf:knows>

<foaf:Person><foaf:name>Victor Duran Levin</foaf:name></foaf:Person>

</foaf:knows></foaf:Person>

Ontology definition• a structure capturing semantic

knowledge about a certain domain by describing relevant concepts and relations between them

• a graph / network structure consisting from:– A set of concepts (vertices in a

graph);– A set of relationships

connecting concepts (directed edges in a graph);

– A set of instances assigned to a particular concepts

Semantic Web Ontology languages

• RDF: the Resource Description Framework (RDF) and RDF Schema

– is essentially a data modeling language

– RDF is graph-based, but usually serialized as XML.

– it consists of triples: subject, predicate, object.

• DAML+OIL: Defense Advanced Research Projects Agency (DARPA) Agent Markup Language (DAML) + Ontology Inference Layer (OIL)

– Fairly comparable to OWL • OWL: Web Ontology Language (OWL)

– the most expressive of the ontology languages

– provides mechanisms for creating all the components of an ontology

Ontology construction1. Defining domain and Scope

– Semantic social networks, trust social networks on semantic web

2. Data understanding– Social meta-data; user profiles

and information of people and their social connections

3. Task definition– Describing and representing trust

relationships4. Ontology learning 5. Ontology evaluation6. Refinement with human in

the loop

Main sphere of trust ontology. 3 main Concepts of trust ontology as well as two relationships connecting them together. (visualized using Ontosphere3D plug-in for ProtégéOntosphere3D plug-in for Protégé.)

Relationship concept; two main properties of hasTrustee And hasTruster defined on the range of Foaf-agent and connected to AuxiliaryProperties And MainProperties using two relations of hasAuxiliaryProperties and hasMainProperties

MainProperties concept; having two main data type properties of trust subject (topical trust), and trust value

AuxiliaryProperties concept; having 3 data type properties of DateBegin, DateEnd and ContextType and also hasRecommender which is defined on the range of foaf agent

Modeling trust• We have considered modeling trust • We have described trust in

performance. When we state that “Alice trusts Bob regarding DrivingAlice trusts Bob regarding Driving”.– meaning that, “Alice trusts in

eventuality of performance of Bob to some extent, when the act of driving is performed”.

• Trust Metric: We have used a probabilistic approach to describe trust relationships, so we can say how much someone trusts the other on a range between 0 and 1.

Modeling distrust and feelings• Distrust, is modeled implicitly.

– For instance,” Alice distrusts Bob regarding babysitting to some extent (0.65)”, can be also stated like “Alice trusts Bob regarding babysitting to some (complementary) extent (0.35)”.

• Feelings is modeled to some extent. – Averaging all evaluation values

for a relation, We can derive negative or positive feelings.

• If Alice has low trust values for Bob, then we can state that she has negative feelings for him, or vice versa.

– Limitation: there are many certain properties that should be considered

Presentation of Network using our ontology

<foaf:Person rdf:ID="Alice"/> <foaf:Person rdf:ID="Bob"/> <Relationship rdf:ID="Relationship_Alice_Bob"> <hasTrustee rdf:resource="#Bob"/> <hasTruster rdf:resource="#Alice"/> <hasMainProperties>

<MainProperties rdf:ID="MainProperties_Alice_Bob"><Subject

rdf:datatype="&xsd;string">Driving</Subject><Value rdf:datatype="&xsd;float">0.95</Value>

</MainProperties> </hasMainProperties> <hasAuxiliaryProperties>

<AuxiliaryProperties rdf:ID="AuxiliaryProperties_Alice_Bob"><ContextType rdf:datatype="&xsd;string">

Social Network</ContextType>

</AuxiliaryProperties> </hasAuxiliaryProperties> </Relationship>

BobAlice

Trust / Driving

0.95

Trust networks of small size

ClaraBob

David

Alice

Trust / Cooking

Distrust / Dishwashing

Trust / Driving

Distrust / Teaching

0.76

0.46

0.96

0.80

Increasing the length of the network structure, Network contains 20 nodes and 34 edges)

ClaraBob

David

Alice

Increasing the width of the network structure, (Network contains 28 nodes and 54 edges)

Trust networks of larger size

ClaraBob

David

Alice

GingerEric

Henrik

Frida

Social Network

Business Network

Hybrid Networks - Connected networks of different contexts; a social and a business network. Example Hybrid networks, contain 8 people and 12 relations. 8 links are interconnections (local), and 4 links are acting as glue connecting two networks (foreign). Network contains 48 nodes and 92 edges.

Trust networks of larger size

CB

D

AaB

GE

H

F

OM

P

N

KI

L

J

Meshed networks –Motivated from real-world network formations, complex, combined networks of different sizes and different contexts. Partial or fully connected meshed networks. In sample, A partial meshed network made-up of two connected hybrid networks; This network contains 16 people and 26 relations. (Network contains 98 nodes and 198 edges)

Structural comparison; Small sized networks

GolbeckOurs

Konfidi

Nodes

Edges0

5

10

15

20

25

30

35

40

Nodes

Edges

Increase in length; Networks of 4 people and 4 relationships.

Increasing is width; Networks of 4 people and 6 relationships.

GolbeckOurs

Konfidi

Nodes

Edges0

10

20

30

40

50

60

Nodes

Edges

Structural comparison; large sized networks

Hybrid Network ; Network of 8 people and 12 relationships.

Meshed network; Networks of 16 people and 26 relationships.

GolbeckOurs

Konfidi

Nodes

Edges0

20

40

60

80

100

120

Nodes

Edges

GolbeckNima

Konfidi

Nodes

Edges0

50

100

150

200

250

Nodes

Edges

Discussion on comparisons• In networks of small size, ontology

shows average performance.– As the size of the networks increase,

certain aspect of trust network size increases

• Main reason, the number of elements incorporated within the structure of ontology. – Golbeck’s ontology uses only one main

element, Konfidi uses two main elements, while our ontology uses three main concepts

• Efficiency in design of the ontology– Efficient design has certain aspects that

reduce the size of the networks generated using ontology (Structural Determinism)• Management of over all organization

Discussion on comparisons• Third reason is the

AuxiliaryProperties – incorporating an extensibility

element, incorporates extra edges and especially extra nodes into the network.

• Important: None of the other compared ontologies, have no elements for describing extra properties; – extending Golbeck’s trust

ontology is very hard and needs a thorough change

– Konfidi doesn’t have any elements for describing extra properties.

Conclusion• We analyzed the modeling and

representation of trust within semantic web-driven networking societies. – We used Ontologies as our

modeling tool.• There are certain new features

that our work introduces to trust ontologies in this context; – using our AuxiliaryProperties, we

give relationships more weight and meaning.

– We have introduced the hasRecommender property that can determine the strength of the links on social network

Conclusion

• As a conclusion, ontologies are very promising technologies.

• Utilizing ontologies in modeling and representing trust in semantic web-enabled social networks seems to be a highly efficient methodology and mechanism

• Maybe, “KEY FREE TRUST” KEY FREE TRUST” AT LAST !AT LAST !

Future work• Spotting more research fields

on overlapping areas between Social sciences and Semantic web

• One of the most important future works is spotting further applications for social trust, where trust relationships can be modeled and expressed using ontologies. – Current applications are just limited

to Spam filtering and user rating systems across web sites on internet.

Questions

• Thank you for your attention !

Nima Dokoohaki