VIeTE
Enabling Trust Emergence in Service-oriented Collaborative Environments
Florian Skopik, Hong-Linh Truong and Schahram Dustdar
Distributed Systems Group
Information Systems Institute
Vienna University of Technology, Austria
Keywords:
Activity-centric collaboration, Trust, Service-oriented architecture.
Abstract:
In activity-centric environments where people from different companies and disciplines work remotely to-
gether and where new virtual teams are formed and dissolved continuously, how to find the most suitable
collaboration partner for a given task and how well one partner is able to collaborate with another one are
challenging research questions. Determining and considering people’s professional competencies, collabora-
tion behavior and relationships is a prerequisite to enhance the overall collaboration performance and success,
because these factors highly impact on the notion of trust used to select and grade partners. In this paper we
analyze these factors and their impact on trust relationships in modern service-oriented collaboration environ-
ments. We present VieTE, a framework for trust emergence therein supporting the analysis of trust between
partners in various contexts and from different views. In contrast to other approaches, which mostly rely on
manual and subjective user feedback, VieTE monitors automatically collaboration efforts and deduces trust
between any two partners based on past collaboration, previous successes, and individual competencies.
1 INTRODUCTION
With the rise of information and communication tech-
nologies, the way people organize and perform their
work has been shifted to a distributed form, where
people located at different sites build loosely coupled
teams and work together to reach common goals. In
such working scenarios, it is difficult for team mem-
bers to establish personal relationships. However, es-
pecially in team oriented environments, one aspect
of interpersonal relationships must not be neglected,
which is trust. High trust between collaboration part-
ners is vital to collaboration processes and thus to
their overall success.
A wide range of systems have been proposed for
establishing trust, such as those described in (Jøsang
et al., 2007), but most of them rely on subjective
user feedback, which is time-consuming for the users
and error-prone due to social influences or malicious
raters. In order to overcome this user feedback de-
pendency and to automate the rating process, we fol-
low a monitoring-based approach by observing and
analyzing users’ communication and behavior to de-
termine notions of trust during collaborations in ad-
hoc, service-oriented collaboration environments. To
this end, we have developed the Vienna Trust Emer-
gence Framework (VieTE) to support the analysis of
trust between any collaboration partners. By deter-
mining and providing trust values directly from mon-
itoring collaborations, VieTE improves the support
for typical use cases in ad-hoc collaboration scenarios
including selecting a partner or service at run-time,
permitting user recommendation and ranking, allow-
ing trust-based team formation, supporting trustwor-
thy resource access control or enabling evaluation of
team performance; just to mention a few examples.
The remainder of this paper is organized as fol-
lows. Section 2 describes our supporting service-
oriented collaboration environment. Section 3
presents the related work. In Section 4 we define trust
together with its context and views. We present data
collection and complex interaction metrics for deter-
mining trust in Section 5. Section 6 describes VieTE’s
architecture and functionalities. We discuss illustra-
tive scenarios in Section 7 and conclude the paper in
Section 8.
471
Skopik F., Truong H. and Dustdar S.
VIeTE - Enabling Trust Emergence in Service-oriented Collaborative Environments.
DOI: 10.5220/0001820904710478
In Proceedings of the Fifth International Conference on Web Information Systems and Technologies (WEBIST 2009), page
ISBN: 978-989-8111-81-4
Copyright
c
2009 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
2 SERVICE-ORIENTED
COLLABORATION
ENVIRONMENT
We consider all motivating use cases mentioned in the
introduction, and describe a service-oriented collab-
oration environment which is generic enough to be
used in a wide range of real scenarios. In this en-
vironment humans are organized in teams perform-
ing activities with the support of SOA-based services.
In this regard the term collaboration means that peo-
ple work together in various ways to reach a common
goal. All tasks they perform are organized in activi-
ties, which are structures to help managing and mon-
itoring which humans are jointly performing which
tasks by utilizing which services. Figure 1 shows an
overview of involved entities and their relationships.
belongs to
interact
within A
performs
is assigned to
interact
within A
subordi
-nates
interact
within A
HumansTeamsActivitiesServices
A
Figure 1: Relationships between entities.
A human is a single person who belongs to one
or more teams at the same time. Every team member
has one or more roles, which describe their respon-
sibilities within a team (e.g. leader, contributor etc.),
interacts with services in context of particular activi-
ties and interacts with other humans of the same team.
Every interaction has a particular purpose, thus we
can distinguish between various types including coor-
dination (e.g. agree on a meeting schedule), commu-
nication (e.g. send instant message or e-mail) or exe-
cution (e.g. execute a service to fulfill a task), depen-
dent on which category of service is utilized. In the
described environment we monitor different kinds of
interactions, such as SOAP-based (Web) service calls,
and e-mail and instant messages.
An activity is any kind of basic task (e.g. testing
a software module) performed during work and is ex-
ecuted by exactly one team, however a team can be
assigned to several activities at the same time. Ac-
tivities are hierarchically structured and can have an
arbitrary number of subactivities. Furthermore, activ-
ities have a particular goal and nature, e.g., in a soft-
ware testing activity a final goal may be the upload of
a test report to a repository. All interactions between
humans and/or services take place in context of par-
ticular activities.
A service is a resource assigned to an activity pro-
viding support for a team during an activity’s exe-
cution. We are operating in a mixed systems envi-
ronment, where services may be commonly known
software Web services or humans acting as services
(e.g., through human provided services (Schall et al.,
2008)). Such an environment permits supporting
more complex tasks, which cannot be tackled by tra-
ditional Web services, but by humans using common
Web service technologies and widely adopted infras-
tructures. If several services are assigned to the same
activitythey can interact with each other, building ser-
vice compositions to offer extended functionalities.
3 RELATED WORK
In this paper we present a framework, VieTE, to en-
able trust emergence in human collaboration environ-
ments, which are similar to previous activity-centric
approaches, such as IBM’s UAM
1
presented for ex-
ample in (Moody et al., 2006). Several projects in the
field of collaboration are currently performed within
the EU FP7
2
, partly utilizing the concept of trust.
From the comprehensive surveys of trust in com-
puter science, including (Jøsang et al., 2007), (Ruo-
homaa and Kutvonen, 2005) and (Artz and Gil, 2007),
we select and extend the trust definitions in (Mui,
2002) and (Grandison and Sloman, 2000) which fit
best to our framework as explained in Section 4.
The overall use of VieTE is related to classic rec-
ommender systems (Resnick and Varian, 1997) and
collaborative filtering approaches (Herlocker et al.,
2004), however,in these fields the opinion of an entire
community about a single entity is predominantly rec-
ognized while relations between single entities within
a group are often neglected. Our work introduces the
concept of views of trust, which allows to grade an
entity from different perspectives, including from an
1
http://www.research.ibm.com/uam/
2
http://cordis.europa.eu/fp7
WEBIST 2009 - 5th International Conference on Web Information Systems and Technologies
472
entire communiy’s view but also from an individual’s
view. There are many reputation models from the
SOA domain, such as (Maximilien and Singh, 2004) ,
but they are dedicated to Web services only.
In contrast to mentioned reputation systems, in the
domain of social network analysis (Wasserman and
Faust, 1994) the relationships between single entities
are highly researched. From this area we get valuable
input about the composition of typical user communi-
ties, such as (Gomez et al., 2008). Experimental case
studies, including (Massa and Avesani, 2005), offer
insights in human collaboration behavior and enables
us to define requirements for our framework and a ba-
sic trust model.
The aim of trust models is to abstract the fuzzy
notion of trust and to build a mathematical model to
enable systematic trust calculation and analysis be-
tween any entities. There are several papers deal-
ing with the definition of trust metrics and models
in general (Ramchurn et al., 2004), (Huynh et al.,
2006), (Theodorakopoulos and Baras, 2006) or fo-
cusing particular aspects such as propagation (Guha
et al., 2004), (Quercia et al., 2007) or mobility (Shand
et al., 2004). For basic trust determination, we utilize
the widely adopted concept of a trust graph where its
nodes represent acting entities and weighted edges re-
flect the relationships among entities.
4 CONTEXT AND VIEWS OF
TRUST
Before discussing the context and views associated
with trust, we define the notion of trust in our frame-
work. Trust has been defined in several different ways
depending on the research area, such as in (Grandison
and Sloman, 2000) and (Artz and Gil, 2007). A defi-
nition of trust from (Mui, 2002) suitable for the intro-
duced collaboration environmentstates that trust is ”a
subjective expectation an agent has about another’s
future behavior based on history of their encounters..
The point is that inferred from previous collaboration
behavior and experiences a notion of trust is deduced.
Particularly in collaboration scenarios where peo-
ple from different companies and from a wide range
of disciplines work together using services from sev-
eral vendors, they are often unknown to each other,
thus trust cannot rely on personal relationships, but
has to be mostly determined by the success of past
collaborations and the quality of the outcome only.
Hence we argue that one’s trust in another one is
higher, the more efficiently both performed in the
same activities and teams respectively. In the de-
scribed collaboration environment success and effi-
ciency basically depend on the competencies of the
acting entities, in detail humans and services. A hu-
man offers competencies such as special skills and ca-
pabilities and a service functionalities and features to
support particular activities. From this perspective,
we adopt the definition from (Grandison and Sloman,
2000), trust is ”the firm belief in the competence of an
entity to act dependably, securely, and reliably within
a specified context”. Based on that, we combine and
extend above definitions of trust to define trust in our
collaboration environment as:
a subjective opinion based on previ-
ous collaboration experiences one entity has
about another’s competencies to act depend-
ably, securely, and reliably within a speci-
fied context, determined by performed activ-
ities and involved teams.
4.1 Context of Trust
It is widely agreed (McKnight and Chervany, 1996),
(Marsh, 1994), (Grandison and Sloman, 2000) and an
integral part of the above definition that trust is con-
text dependent, which means it is determined for par-
ticular entities in particular situations. In contrast to
a wide range of reputation systems (Herlocker et al.,
2004), which calculate only one kind of globally valid
trust value, independent from situation and use case,
we derive trust for certain humans and services with
respect to their context. Particularly in the introduced
collaboration environment context, which reflects a
real situation, can be fully described by the notions
of team and activity. The team holds structural in-
formation about which humans work closely together
to reach a common goal and which roles they have,
while the activity describes the goal itself to reach.
Thus, contextual information of an entity includes all
properties available about current and past activities
and teams involved.
4.2 Views of Trust
As mentioned in the related work section, trust has
been widely defined (i) from an individual view in so-
cial networks, where relations between single users
are maintained or (ii) from a global view in reputa-
tion systems or collaborative tagging systems, which
mostly use an aggregation of the individuals’ views.
For typical collaboration scenarios, where humans are
tightly coupled and form teams, we introduce one
level in between by taking a team’s view of trust into
account. This enables us to determine trust of an en-
tire team into another entity, which is a basic demand
VIeTE - Enabling Trust Emergence in Service-oriented Collaborative Environments
473
in collaborativedecisions, such as the selection of fur-
ther team members or services.
In the described collaboration environment we
distinguish between the following views of trust,
which can be created by combining contextual infor-
mation from different individuals:
individual view: describes trust of one human in
another one or in a service.
team view: describes trust of an entire team in
one human or service. For determining team trust
previous collaboration encounters from all team
members with a particular entity are aggregated.
global view: determines trust in a human or ser-
vice from a global point of view, similar to global
reputation systems, where all available informa-
tion within a collaboration scenario is taken into
account.
Figure 2 shows examples of the three views of
trust in a service and lists the influencing factors for
trust determination.
Individual View
Team View
Global View
Determined by:
- aggregated team views
- further available information
(e.g. QoS attributes)
Determined by:
- individual view of each mem-
ber and their relationships
- each member’s role and
participation in the team
- team’s previous successes
Determined by:
- individual experience with
the service derived from
previous interactions
- experience, derived from the
opinions of well-known colla-
boration partners
Service
Service
Service
Figure 2: Distinguished views of trust in a service.
The differentiation into diverse views of trust com-
bined with the notion of trust context is the basis for
a comprehensive customization of our framework (as
further described in Section 6).
5 COLLECTING DATA FOR
DETERMINING TRUST
To enable VieTE to provide notions of trust from dif-
ferent views and in different contexts, we need an
appropriate data model supporting the described col-
laboration environment, and utilize a number of data
sources to determine trust.
5.1 Data Model
The four main entities of the data model human, team,
activity and service, and their connections as men-
tioned in Section 2, are modeled in Figure 3. Ev-
ery entity is described with further attributes, their
so called profile data, serving as input for trust de-
termination. These profiles describe a collaboration
scenario including its participating entities, and are
available to all entities within the same collaboration
scenario. For example all humans within a team share
their profile data with each other.
Figure 3 depicts our data model. We distinguish
two different kinds of data, which are (i) profile data,
describing an entity’s properties and structural re-
lationships and (ii) interaction data, describing dy-
namic collaboration encounters.
The data model is designed to be generic enough
to handle different forms of collaboration scenarios
with a wide variety of characteristics. If there is, for
instance, no need for a notion of team, all humans can
be assigned to a single team or each human may repre-
sent an own team respectively. The entity’s properties
are selected to be applicable in various contexts.
5.2 Data Sources
Besides information about human profiles and team
structures, we store data about current and past activ-
ities as well as service profiles. Furthermore, in our
approach we do not rely on subjective user feedback
or reputation, for instance, in form of questionnaires,
but we take into account what can be measured auto-
matically. We use a Web services infrastructure which
enables us to log low level interaction messages by
using Web service handlers and SOAP interceptors.
Hence, we are able to capture (i) human-human in-
teractions as long as they take place via observable
communication services (ii) human-service interac-
tions, (iii) service-service interactions in service com-
positions, (iv) predefined service or application events
in customized logging scenarios (e.g. e-mail traffic
over SMTP or SVN file accesses), and (v) changes of
team- and activity structures including adding team
members, changing roles or assigning new services to
an activity.
For this measurement approach it is necessary
to determine the success of interactions. Depend-
ing on the type of interactions and participating en-
tities, interaction failures can have manifold reasons,
WEBIST 2009 - 5th International Conference on Web Information Systems and Technologies
474
Human
id job position
date of hire educational level
Service
id vendor
date of last update category
Team
id humans[ ] with roles[ ]
date of formation size
date of dissolving
Activity
id type
startdate goal
duration assigned team
status/progress assigned services[ ]
1
belongs to
performs
is assigned to
Interaction
success time
type number of participants
1
1 1
Interaction
success time
type
Interaction
success time
type
Figure 3: Simplified data model of profiles and interactions.
such as SOAP exceptions if a particular service is
down, application specific errors due to missing fea-
tures or wrong usage of services, not running in-
stant messengers or e-mail clients, unanswered in-
stant messages, missed read notifications of e-mails,
interrupted data transfers, invalid communication set-
tings, missed phone calls etc.
5.3 Collaboration Metrics
Data from all mentioned sources is captured and ag-
gregated to more detailed composite metrics and pat-
terns providing meaningful information for a collab-
oration scenario. The calculation process is fully cus-
tomizable and can be set up for different views of
trust.
Table 1: Exemplary metrics for humans.
View Metric Description
Indiv. scores for initiated, accepted, successful and failed
human-to-human/service interactions.
interaction success with a particular human/service.
experience with a particular activity type.
Team human’s average impact on team performance.
human’s interaction participation within a team.
Global metrics for team view can be applied too, but aggre-
gated over all teams.
Table 2: Exemplary metrics for teams.
View Metric Description
Indiv. only profile data is available.
Team average number of members’ interactions with a par-
ticular entity.
interaction distribution among members.
service access distribution (with respect to a particu-
lar or all activities).
Global amount of interactions compared to other teams.
activity participation compared to other teams.
team success compared to other teams.
We elaborate some exemplary metrics in Tables 1
to 4 to show the potential of our approach. The exact
Table 3: Exemplary metrics for activities.
View Metric Description
Indiv. none. Activities are performed by entire teams.
Team collaboration effort compared to other activities.
success of service usage within an activity.
success of executive team within an activity.
success of artifact outcome.
service access distribution within an activity.
activity contribution of every team member.
Global metrics for team view can be applied too, but com-
pared to activities of all teams.
Table 4: Exemplary metrics for services.
View Metric Description
Indiv. scores for initiated, accepted, successful and failed
service-to-service/human interactions.
interaction success with a particular human/service.
Team average number of invocations.
usage factor by a particular team.
service access distribution by a particular team.
Global same metrics as for team view aggregated over all
teams.
definitions of these metrics depend on available col-
laboration data and thus on the environment, and is
not in scope of this paper, which focuses a high-level
overview of the whole approach. In complex collab-
oration scenarios more metrics, especially aggregated
from simpler ones, are possible. Currently we focus
on basic metrics which mostly depend on interaction
data, such as interaction scores and success rates.
6 VIeTE ARCHITECTURE
To support trust in the described collaboration envi-
ronment VieTE consists of (i) tools for managing hu-
mans, teams, activities and services, utilizing services
in the back-end, (ii) sensors and logging mechanisms
to monitor interactions at run-time, and (iii) a trust
determination service to deal with activity structures,
VIeTE - Enabling Trust Emergence in Service-oriented Collaborative Environments
475
service information, human and team profiles, and in-
teraction logs, to calculate metrics, and ultimately to
determine trust between any interacting entities. Fig-
ure 4 depicts the architecture of VieTE.
6.1 Determining Trust
In collaboration scenarios, humans use several tools
to manage their activities, to search for suitable part-
ners or services and to formate teams. VieTE pro-
vides trust-aware support for these tasks. To this end,
we monitor collaborations of humans with other hu-
mans and services to collect data for trust determina-
tion. We describe the mainly involved components in
the following:
Interaction Sensors and Logging. Interactions
between and among humans and services are captured
by sensors. We have developed mechanisms to in-
tercept SOAP calls to services, e-mail traffic, instant
messages over XMPP
3
and SVN document repository
accesses, however this can be extended by sensors for
a various range of open or proprietary communica-
tion protocols, including voice calls or file sharing.
The logged entries contain at least the type of interac-
tion, sender-id, receiver-id and a timestamp; further-
more, dependent on interacting entities, the endpoint
interface, invoked operation, parametersetc. All com-
munication and application errors are logged as well,
e.g., SOAP exceptions or access denials.
Collaboration Metrics Calculation. Based on
collected interaction data and information from col-
laboration management services this component cal-
culates context dependent metrics for humans, teams,
services and activities, as described before.
Trust Model. This component implements a
directed graph, where the nodes represent humans
and services and the links reflect their trust relation-
ships. These relationships depend on the metrics cal-
culated in discrete time steps according to the con-
text described by activities and teams. The realization
of these functionalities depends on the utilized trust
model, but currently we build trust using weighted
averages of preselected metrics. The model is cus-
tomized by user specified policies, which control the
metrics to be used and how these metrics are com-
bined, i.e. the weighting factors, to deduce a notion
of trust.
Trust Provider and API. The provider extracts
data from the trust model and creates context depen-
dent views according to API parameters. Collabo-
ration Management Tools can access the trust model
data through the Trust API. The following excerpt of
3
http://www.xmpp.org/
this interface shows the signature of the methods pro-
vided to obtain the trust relationship from an entity
trustor
to an entity
trustee
restricted to a context
ctx
. The view of trust is derived from the type of the
entity trustor, which may be either individual, team or
global.
int getTrust(Entity trustor, Entity trustee, Ctx ctx);
List<Entity> getTrustors(Entity trustee, Ctx ctx);
List<Entity> getTrustees(Entity trustor, Ctx ctx);
...
6.2 Prototype Implementation
The VieTE portal consists of a Liferay Enterprise
Open Source Portal
4
in which tools run as JSR-168
5
compatible portlets. We use portlets for visualizing
and processing user inputs only, while the main part
of the business logic is encapsulated in Web services.
In the back-end, we use a Tomcat server with a
deployed Axis2 for hosting Web services. We utilize
existing, earlier developed services for registering and
managing humans, teams, activities and services, and
for interaction logging; and develop new services for
all tasks concerning trust determination and manage-
ment. Communication between portlets and services
is realized with a SOAP based Web service stack. All
relevant data is stored in a IBM DB2.
For the implementation of the trust model and ap-
plying basic graph algorithms we utilize the Java Uni-
versal Network/Graph Framework
6
.
7 ILLUSTRATIVE SCENARIOS
We set up a collaboration scenario in the field of soft-
ware development which consists of two teams as
shown in Figure 5, where trust values [0, 1], rang-
ing from no to full trust, are calculated between any
two entities, humans or services which directly in-
teract. These values are derived from the types and
amount of successful interactions compared to the to-
tal amount of interactions between two particular en-
tities.
For the sake of simplicity, we assume each team
is predominantly involved in only one type of activ-
ity, which is software implementation for team
A
=
{H
1
, H
2
, H
3
, H
8
} and software testing for team
B
=
{H
4
, H
5
, H
6
, H
7
, H
8
}. Each human has a particular
role (developer, assistant, trainee) in a team. H
8
is
member of both teams, thus has two roles. For the
4
http://www.liferay.com
5
http://jcp.org/aboutJava/communityprocess/final/jsr168/
6
http://jung.sourceforge.net/
WEBIST 2009 - 5th International Conference on Web Information Systems and Technologies
476
Service
Interaction
Scenario
Collaboration Metrics Calculation
Activity
Management
Service
Management
Interaction
Logging
Human/Team
Management
Trust Model
Trust Provider
Trust Query API
Service
Service
Service
Team A
Team B
Trust Determination
Service
interaction
event
Partner
Search
Team
Formation
Activity
Management
Service
Search
trust enabled
search
traditional human/team/service/activity
search and management
Portal Tools for Collaboration Management
Collaboration Management Services Monitoring
humans using
portal tools
activity
structures
interaction
logs
service
info
services supporting
collaboration management
human and
team profiles
Figure 4: Architectural overview of the framework.
Activities of Team
A
Activities of Team
B
(software implementation) (software testing)
0.9
0.6
0.6
0.8
0.9
0.9
0.6
0.8
0.6
0.2
1.0
0.6
0.6
0.9
0.9
0.6
1.0
0.9
0.2
0.4
0.3
0.3
0.6
?
Code Generation
Service S
1
H
8
(dev,15%,
dev,10%)
H
3
(trainee,5%)
H
1
(dev,60%)
H
5
(trainee,5%)
H
2
(dev,20%)
H
4
(dev,20%)
H
7
(dev,50%)
H
6
(assistant,15%)
0.2
Figure 5: Experimental setup with humans having different
roles and interaction participation within teams (in %).
creation of global and team views we built the aver-
age of the involved individuals’ trust values weighted
by their interaction participation. Interaction partici-
pation refers to the amount of successful interactions
performed by one human expressed in percent of all
successful interactions within a team.
Scenario 1: Trust Determination. H
8
wants to
know if it is worth using service S
1
, however, because
H
8
never used it in the past, H
8
has to rely on others’
notions of trust. Table 5 shows the calculated trust
values of different trustors according to their contexts.
Which one is best applicable depends on the purpose
for which H
8
intends to use S
1
. If H
8
wants to uti-
lize S
1
for software implementation, relying more on
team
A
s view, especially the view of its developers is
wiser. For software testing the situation is different
and S
1
is less trusted than for implementation activi-
ties. Furthermore, it is obvious that trainees trust S
1
less than developers, because high experience is re-
quired to operate S
1
properly and thus S
1
has not been
used with high success by unexperienced trainees in
the past.
Table 5: Different trustors’ trust values in service S1 de-
pending on view and context.
Trustor View Contextual Restrictions Value
all global 0.667
all global activity.type=swimpl. 0.859
all global activity.type=swtest 0.489
all devs global human.role=dev 0.733
all trainees global human.role=trainee 0.25
teamA team 0.859
teamAs devs team human.role=dev 0.9
H
8
indiv. unknown
Scenario 2: Partner Recommendation. In this
scenario we use VieTE to find suitable collabora-
tion partners for H
8
. For the global and team views
the trust relationships of all entities and all members
within a team respectively are considered, as in com-
mon reputation systems, and the results are not ded-
icated to H
8
only. However, for the individual view
only direct relationships from H
8
to others are con-
sidered.
Table 6 shows a list of trustees, ordered descend-
ing by trust values calculated for particular contexts.
From the global view result generally well trusted
partners, however from H
8
s individual view VieTE
suggests only partners, with which H
8
personally in-
VIeTE - Enabling Trust Emergence in Service-oriented Collaborative Environments
477
Table 6: H
8
s trustees depending on view and context.
View Contextual Restriction Trustees
global H4,H7,H1,H2,H6,H3,H5
global activity.type=swimpl. H1,H2,H3
global human.role=dev H4,H7,H1,H2
team human.team=team
A
H1,H2,H3
indiv. human.team=team
A
H2,H1,H3
teracted in the past, while others are out of scope.
Note that the order of trustees for H
8
s individual view
and team
A
s team view are different, though contex-
tual restrictions are set to take into account only hu-
mans from team
A
in both cases. This is due to the
fact that H
8
s trust in H
2
is quite high (0.9), while the
trust of other team members in H
2
is only medium
(0.6 on average), thus on team
A
s view H
2
is ranked
lower than H
1
.
8 CONCLUSIONS AND FUTURE
WORK
In this paper we have discussed the role of trust and
related concepts in service-oriented collaboration en-
vironments. We defined a collaboration model com-
prising of humans and services, and proposed an ap-
proach and a framework to automate trust determi-
nation based on monitoring interactions and utilizing
profile information.
Currently we focus on trust metrics and models
and extend the existing prototype to make it feasible
for supporting real world scenarios in the area of net-
worked enterprises. The challenge is the definition
of suitable metrics which are able to reflect real trust
relationships. After that, the next step will be to per-
form an empirical evaluation and to prove that the se-
lected metrics and models appropriately address the
challenges in automatic trust inference.
ACKNOWLEDGEMENTS
This work is mainly supported by the EuropeanUnion
through the IP project COIN (FP7-216256).
REFERENCES
Artz, D. and Gil, Y. (2007). A survey of trust in computer
science and the semantic web. J. Web Sem., 5(2):58–
71.
Gomez, V., Kaltenbrunner, A., and Lopez, V. (2008). Sta-
tistical analysis of the social network and discussion
threads in slashdot. In WWW, pages 645–654. ACM.
Grandison, T. and Sloman, M. (2000). A survey of trust in
internet applications. IEEE Communications Surveys
and Tutorials, 3(4).
Guha, R., Kumar, R., Raghavan, P., and Tomkins, A.
(2004). Propagation of trust and distrust. In WWW,
pages 403–412, New York, NY, USA. ACM.
Herlocker, J. L., Konstan, J. A., Terveen, L. G., and Riedl,
J. T. (2004). Evaluating collaborative filtering recom-
mender systems. ACM Trans. Inf. Syst., 22(1):5–53.
Huynh, T. D., Jennings, N. R., and Shadbolt, N. R. (2006).
An integrated trust and reputation model for open
multi-agent systems. Autonomous Agents and Multi-
Agent Systems, 13(2):119–154.
Jøsang, A., Ismail, R., and Boyd, C. (2007). A survey of
trust and reputation systems for online service provi-
sion. Decision Support Systems, 43(2):618–644.
Marsh, S. P. (1994). Formalising trust as a computational
concept. PhD thesis, University of Stirling.
Massa, P. and Avesani, P. (2005). Controversial users de-
mand local trust metrics: An experimental study on
epinions.com community. In AAAI, pages 121–126.
Maximilien, E. M. and Singh, M. P. (2004). Toward au-
tonomic web services trust and selection. In ICSOC,
pages 212–221. ACM.
McKnight, D. H. and Chervany, N.L. (1996). The meanings
of trust. Technical report.
Moody, P., Gruen, D., Muller, M. J., Tang, J. C., and Moran,
T. P. (2006). Business activity patterns: A new model
for collaborative business applications. IBM Systems
Journal, 45(4):683–694.
Mui, L. (2002). Computational models of trust and rep-
utation: Agents, evolutionary games, and social net-
works. PhD thesis, Massachusetts Institute of Tech-
nology.
Quercia, D., Hailes, S., and Capra, L. (2007). Lightweight
distributed trust propagation. In ICDM, pages 282–
291. IEEE Computer Society.
Ramchurn, S. D., Jennings, N. R., Sierra, C., and Godo,
L. (2004). Devising a trust model for multi-agent in-
teractions using confidence and reputation. Applied
Artificial Intelligence, 18(9-10):833–852.
Resnick, P. and Varian, H. R. (1997). Recommender sys-
tems. Communications of the ACM, 40(3):56–58.
Ruohomaa, S. and Kutvonen, L. (2005). Trust management
survey. In iTrust, volume 3477 of Lecture Notes in
Computer Science, pages 77–92. Springer.
Schall, D., Truong, H.-L., and Dustdar, S. (2008). Unifying
human and software services in web-scale collabora-
tions. IEEE Internet Computing, 12(3):62–68.
Shand, B., Dimmock, N., and Bacon, J. (2004). Trust for
ubiquitous, transparent collaboration. Wireless Net-
works, 10(6):711–721.
Theodorakopoulos, G. and Baras, J. S. (2006). On trust
models and trust evaluation metrics for ad hoc net-
works. IEEE Journal on Selected Areas in Commu-
nications, 24(2):318–328.
Wasserman, S. and Faust, K. (1994). Social network analy-
sis: Methods and applications. Cambridge University
Press.
WEBIST 2009 - 5th International Conference on Web Information Systems and Technologies
478