Eymen Ben Bnina, Olivier Camp, Chùng Ti
ên Nguy
Department of Computer Science, ESEO, Angers, France
Hella Kaffel Ben Ayed
Faculty of Science, Tunis, Tunisia
Trust models, context, ad hoc networks.
The global performance of a mobile ad hoc networks (manet) greatly depends on, both, the cooperation of
participating nodes and the environment in which the nodes evolve. The willingness of a node to cooperate can
be illustrated by the trust assessed to the node. Yet, existing trust models, designed for reliable wired networks,
do not take into consideration possible communication failures between client and server. We believe, that in
the case of ad hoc networks such factors should be considered when computing trust. In this article, we show
how an interaction can be decomoposed in three separate phases : two communication phases for transporting
the request to the server and the response back to the client, and one execution phase which represents the
actual execution of the service by the server. We propose to define the communication environment using
contextual attributes and to consider this context when assessing trust to a server. We discuss the possible uses
of context in the field of trust computation in manets and define contextual attributes that seem important to
consider when modelling and computing trust.
A distributed system is a system consisting of nodes
which cooperate together provide users with services
such as web services, data exchange and sharing of
software. Such an environment is said to be fully dis-
tributed if there is no central component controlling
the whole system. Besides centralised distributed sys-
tems, in which one or more central servers control the
entire system, we can find fully distributed systems in
which all nodes are equals and are, together, in charge
of controlling the system. Mobile ad hoc networks
(manets), also called spontaneous networks, are an
example of such fully distributed systems.
Manets are IP networks made up of a collection
of mobile nodes communicating via radio links. They
do not rely on any predefined infrastructure or cen-
tralised administration to operate. Nodes in a manet
may, at any time, leave, enter or roam within the net-
work. The resulting dynamic nature of the network’s
topology, along with the unreliability of the wireless
links, thus require for manet services to be highly
adaptable. For instance, in the case of routing, the
lack of a network infrastructure implies that the ser-
vice is provided in a peer-to-peer fashion and that all
the nodes need to act as collaborating routers, to pro-
vide multi-hop routes between any source and des-
tination. Moreover, the availability of an individual
central node cannot be guaranteed at all times. There-
fore, services cannot rely on a point of centralisation
and should be provided in a distributed and adaptive
Manets provide an easier way, in comparison with
classical, infrastucture-based networks, to aggregate
large amounts of resources while maintaining a low
system maintenance cost and are thus an interesting
solution when setting up dynamic and flexible appli-
cations. However, these systems also bring up some
new problems. Indeed, in such an environment, when
a node needs to use a given service, it may not know
with which quality of service each server is able to
provide the service. This is particularly true in an
open network in which a node does not knowto which
extent the other nodes are willing to cooperate.
One among the solutions that have been suggested
for such problems, is to control whether a node is, or
not, authorised to enter the network. This way, it can
be considered that the nodes in the network have all
the approval of a trusted central entity and may also
themselves be trusted; according to a policy defined
Ben Bnina E., Camp O., Tiê
n Nguyên C. and Kaffel Ben Ayed H. (2008).
In Proceedings of the Tenth International Conference on Enterprise Information Systems - SAIC, pages 133-141
DOI: 10.5220/0001701001330141
by the trusted central entity. Yet, such limitation on
the number of cooperating nodes may greatly degrade
the performances of the network. Another solution is
to calculate the trust that the client assigns to each
server and, based on this, decide whether or not to
cooperate with the service provider. Several works
have tackled the problem of computational trust and
different models have been proposed.
The existing trust models are well adapted to sta-
ble and reliable networks, but most are not suitable
for the inherently dynamic nature of manets. Indeed,
it is no point determining the most trustworthy server
for a given service, if the unreliability of the network
does not allow communication with this server. For
this reason, we believe that the factors influencing the
stability of the network should also be considered dur-
ing the evaluation of trust.
In this paper,we present a brief review of some ex-
isting trust models and discuss how researchers have
proposed different approaches for computing trust.
We stress the fact that most models do not explicitely
take situational information into account when com-
puting trust and explain why this is penalising in un-
reliable and dynamic contexts such as manets. Thus,
we discuss the notion of context and present how it
can be used in manets during trust computation.
The remain of this paper is organised as follows:
section 2 gives a brief survey of some existing trust
models, and points out approaches proposed by re-
searchers for computing trust. Section 3 presents the
reasons for which, in our opinion, context should be
considered when computing trust in manets and gives
an overview of existing definitions for context. Sec-
tion 4 gives our perception of context in manets and
proposes a set of contextual attributes that, we think,
should be considered when studying the result of past
interactions when computing trust. We conclude in
section 5 and present our future works.
Trust is a concept which is frequently used in our
daily lives. In fact, it often guides the social inter-
actions with other individuals. Trust has been stud-
ied in various fields (philosophy, economics, sociol-
ogy, psychology and, more recently, computer sci-
ence) thus leading to the existence of several defi-
nitions (Duma et al., 2005)(Grandison, 2003) and a
lack of coherence among researchers. (McKnight and
Chervany., 1996). For us, the most appropriate def-
inition is that proposed by Gambetta in (Gambetta,
1988): "Trust (or symmetrically, distrust) is a partic-
ular level of the subjective probability with which an
agent assesses that another agent or group of agents
will perform a particular action, both before he can
monitor such action (or independently of his capacity
ever to be able to monitor it) and in a context in which
it affects its own action". This definition stresses the
facts that: trust is uncertain, trust is subjective and
trust depends on context.
Several trust models have been described in lit-
erature. The differences between them mainly con-
cern : trust modeling, trust management and decision
making. While trust modeling deals with the repre-
sentational and computational aspects of trust values,
trust management focuses on the collection of evi-
dence and risk evaluation.Concerning decision mak-
ing it is actually a part of trust management. In this
survey, we will concentrate only on trust modeling.
The model defined in (Marsh, 1994) is based only
on direct experiences between trustor and trustee. In
this model, the author introduces three types of trust :
Dispositional Trust T
refers to the trust of an
agent x regardless of the possible cooperation
partner and the situation.
General Trust T
(y) describes the trust of x on y
regardless of the situation.
Situational Trust T
(y, β) describes the trust of
agent x on agent y in situation β. The values of
trust belong to [1, 1[. These trust values illus-
trate the fact that full distrust exists but not total
Abdul-Rahman and Hailes (Abdul-Rahman and
Hailes, 2000) present a trust model based on the so-
ciological characteristics of trust. Their model sup-
ports the following properties of social trust: a) Trust
is context-dependent. b) Trust supports negative and
positive degrees of belief of an agent’s trustworthi-
ness. c) Trust is based on prior experiences. Agents
are able to identify repeated experiences with sim-
ilar contexts and with the same agents. d) Rep-
utational information is exchanged between agents
through recommendations. e) Trust is not transitive -
all evaluations of recommendations take into account
the source of the recommendation. f) Trust is sub-
jective -different agents may have different percep-
tions of the same agent’s trustworthiness. g) Trust
is dynamic and non-monotonic -further experiences
and recommendations increase or decrease the level
of trust in another agent. h) Only Interpersonal Trust
(the trust one agent has in another agent directly in a
specific context) is supported.
This model uses direct trust and recommendor
trust for computing the global trust value assigned by
an agent to the trustworthiness of another. The di-
rect trust that a given trustor agent assigns to another
ICEIS 2008 - International Conference on Enterprise Information Systems
trustee agent A, relatively to a given contextC, is rep-
resented by t(A, c,td) where td is the degree of trust
assigned to agent A and may take one of the following
four values: very trustworthy”, “trustworthy”, “un-
trustworthy”, “very untrustworthy”. In this model,
recommendor trust represents the belief of agent A
concerning the fact that agent B is trustworthy, to a
certain degree, for giving recommendations concern-
ing other agents relatively to a given context c. Rec-
ommendor trust is represented by tr(B, c, rtd) where
rtd is a semantic distance between As perception and
the recommendations it received.
In order to compute the trust that an agent A as-
signs to another agent B, the authors in (Castelfranchi
and Falcone, 1998) propose a trust model based on a
cognitive approach. They assert that the reasons that
make A ask for a service from B result from a set of
mental beliefs. For that, their model is based on the
following three beliefs: 1) Competence belief: A be-
lieves that B can actually do the task; 2)Dependence
belief: A believes that B is the necessary or the best
agent to rely on for doing this task; 3)Disposition be-
lief: A believes that B actually will do the task. The
latter belief is articulated by two other beliefs which
are: i)Willingness belief: A believes that B has de-
cided and will perform action α that is related to the
task; ii)Persistence belief: A believes that B is stable
in its intentions of doing α.
Thus, trust in this model is represented as a set of
mental attitudes which allow agent A to believe that
another agent B will respond to its requests.
The authors in (Yu and Singh, 2001) present a
model which doesn’t combine information relative to
direct and indirect interactions. Recommendations
are only used if an agent has never interacted directly
with the agent for which trust is computed. An ex-
ample of such a situation would be if the trustee has
only recently joined the network. Each node stores
the information concerning direct interactions as a set
of values that reflect the quality of these interactions.
This model only considers the most recent experi-
ences and defines an upper and lower thresholds that
define the limit between what are considered QoS as-
cribed to trustworthy agents, QoS with no clear clas-
sification and QoS ascribed to non trustworthy agents.
By applying the Dempster-Shafer theory of evidence
on the stored information an agent is able to compute
the probability with which the trustee belongs to one
of the above three groups.
In (Nguyen and Camp, 2007), the authors propose
a model that uses, both, direct experiences and rec-
ommendations. They are given as input to a Bayesian
network, and the computing of trust values is based
on a probabilistic approach using Bayes’ theorem of
conditional probability. This model considers that the
future behavior of an agent depends on its past be-
haviors and bayesian networks are very efficient for
manipulating the associated conditional probabilities.
In this model, the trust assigned by agent A to a
server S for providing service S with a quality q is
defined as the probability that S will provide S to A
with the given quality q and is calculated consider-
ing the results of the past interactions with service S
provided by S.
In this section, we have reviewed some of the ex-
isting trust models and pointed out approaches that re-
searchers rely on for computing trust. However these
models do not use situational information which we
believe very important in dynamic environments. In
the following section, we will discuss why contextual
information, specifically in the case of manets, is im-
portant when computing trust. We will also show how
such information may change the trust an agent has in
a server for completing a given task.
For their computing needs, agents in a manet often
have to rely on services provided by others. The
fully distributed nature of a manet does not allow for
a trusted central entity to manage these interactions
and this thus has to be taken in charge of by each
agent. The server providing the required service may
be out of the client’s emission range. In such a case
the client must rely on intermediate nodes to trans-
mit its request, and the server should also use a hop
by hop approach to reply to the client. For the client
to consider its interaction successful, the following
three steps must be accomplished successfully : The
request reaches the server, the server executes the re-
quest, the server’s response reaches the client. This is
represented in figure 1.
On the contrary, if the interaction fails, client C
cannot determine which one of the steps, detailed
Server (S)
Client (C)
3) Response
1) Request
2) Execution
Figure 1: Interaction between agents in a manet.
above, has failed. Moreover,C does not have a global
vision of the network and thus does not know the
route taken by its request, nor does it know the in-
termediate nodes that helped in the transmission of
the response. In fact, the only information the client
holds concerning the routes to and from the server are
those found in its own routing table - ie ; the first hop
and the number of hops to destination. This, of course
is only true if we do not search for such information
in the implementation dependent data manipulated by
the routing protocol. We choose to only consider the
routing table to stay independent from any specific
routing protocol.
According to Gambetta’s definition of trust, the
trust C has that it will obtain a reply to its request
to S depends on: the trust it has that the request will
reach S, the trust it has that S will correctly execute
the service and the trust it has that the reply will be
transmitted back.
Most of the existing trust models only concentrate
on the value of trust assigned to S in properly exe-
cuting the service and choose a collaborator among
the servers with the highest such value of trust. We
believe that such an approach is insufficient, in the
case of manets, and that the correct transmission of
request and response should also be considered. How-
ever, due to the lack of information available con-
cerning the exact participants and their behaviours in
both these transmissions, we consider that an expe-
rience/recommendation based trust model is not ad-
equate for the routing service. Instead, we prefer
to consider that the proper transport of request and
response both depend on the environment through
which those messages are exchanged and propose to
consider context relevant information when comput-
ing trust and deciding of a service provider.
Many definitions have been given for the concept
of context. While in (Salber et al., 1999) authors qual-
ify context as an environment or situation, many re-
searchers use the definition given in (Abowd et al.,
1999): ”Any information that can be used to charac-
terise the situation of entities (i.e whether a person,
place or object) that are considered relevant to the in-
teraction between a user and an application, including
the user and the application themselves”. In (Schilit
et al., 1994), the authors propose to organise the con-
cept of context in three categories:
User Context: user profile, location, people
nearby, social situation, activity, health condi-
tions, agenda settings, etc.;
Execution Context: network traffic, status of the
device, availability of resources, communication
costs, nearby resources, etc.;
Environment Context: weather, light, noise
level, temperature, time, etc.
This categorisation fits well with our vision of an in-
teraction in a manet. Indeed, the routing context influ-
encing the data exchanges between client and server
clearly belongs to the environment context. Also, the
execution context reflects situational factors that may
influence the actual execution of the service by the
server. As for user context, we will see that it can be
used to describe the user profiles of both partners of
the interaction.
The trust models presented in (Castelfranchi and
Falcone, 1998), (Marsh, 1994), (Yu and Singh, 2001)
and (Nguyen and Camp, 2007) are not context aware
and compute trust values independently from context.
Concerning the model presented in (Abdul-Rahman
and Hailes, 2000), authors consider trust in the pres-
ence of virtual communities and leave context man-
agement open to let developers define their own con-
text. In our opinion, contextual information should be
recorded together with every experience and consid-
ered during the trust computation phase. In this work,
we point out the contextual informations that, in our
opinion, should be considered and discuss on how the
context can be used when computing trust in ad hoc
This section has presented the main reason for
which we consider a trust model for manets should be
context-aware and has briefly presented works defin-
ing the concept of context. We must now isolate
the contextual attributes that may be interesting when
considering the mobile, unstable and open environ-
ment of manets.
Even though, to the best of our knowledge, few trust
models take into account the routing context (Toivo-
nen et al., 2006), some consider basic contextual in-
formation when collecting experiences (direct or con-
tained in recommendations). For instance, the model
proposed in (Nguyen and Camp, 2007) records the
time at which each experience occurs and gives less
consideration to older interactions than it does to
more recent ones. We believe this information is
particularly relevant in the case of a manet in which
agents can have varying efficiencies according to their
battery power or even suddenly disappear from the
Another contextual attribute, usually implicitely,
ICEIS 2008 - International Conference on Enterprise Information Systems
taken in charge by some models is the number of
interactions that have occurred between client and
server when trust in the server is calculated. Indeed,
the trust values computed by probability based mod-
els gain precision as the number of interactions with
the server increases. Now, the number of interactions
could be considered as an information relative to the
execution of a service by a given server and thus be
part of the execution context; This is a first step to-
wards context awareness.
Even though the above two contextual attributes
seem important, they have no influence on the trans-
mission of messages between client and server. We
should now decide which contextual attributes are rel-
evant to these steps of the interaction.
The routing context associated to an experience
should contain representationsfor the factors that may
influence the quality of exchanges between client and
4.1 Hop Count
Works presented in (Hekmat and Mieghem, 2003)
show that the probability of success of a communi-
cation between agents is highly dependent on the dis-
tance, in terms of hop count, between the participants;
the higher the hop count, the lower the probability of
success. For this reason we consider the hop count be-
tween client and server to be a crucial information for
caracterising the context of an interaction. This infor-
mation is freely available in the routing table, for any
node with which a communication is possible. How-
ever, the client can only retrieve, the distance to the
server, from its routing table; It cannot determine the
length of the return trip taken by the server’s response.
Indeed, existing routing algorithms, do not necessar-
ily use the same route from one node to another, as
they do than they do when returning. Nevertheless,
we can assume that the return trip is similar, in terms
of number of hops and therefore that the length of the
route from the client to the server also characterises
the travelling of the response.
From our point of view, hop count between client
and server is the first contextual information to be
considered in manets.
4.2 Mobility
The performance in terms of communication, of a
manet is closely related to the capacity of its rout-
ing algorithm to adapt to the mobility of the nodes.
Nevertheless, however efficient routing is in dealing
with mobility, communication between very mobile
nodes, or through a very dynamic network will be less
reliable than between stationnary nodes, or in a sta-
ble environment. Finding out a metric for mobility in
manets is a challenge. There exists few researchs that
have focused on such a topic. In (Boleng et al., 2002),
the authors have used link stability as a metric for mo-
bility to show that mobility has a direct effect on end
to end delay and data packet deliveryratio. Authors in
(Ghassemian et al., 2005) have used a similar metric,
deduced from the frequency of link state change and
link connectivity duration. It thus seems natural to in-
clude information concerning mobility in the routing
context. Moreover, intuitively, it seems to us gener-
ally more efficient for an agent to choose the most
stable server in order to have the greatest probability
that its requets reach the destination server.
What is mobility, how can it be defined and mea-
sured? Is it useful to consider the mobility of the en-
tire network or should we only study that of the path
between client and server? All these questions should
be answered before considering mobility as a part of
the routing context.
Mobility in ad hoc networks is a topic of research
that has been considered through various angles : the
specification of mobility aware routing protocols, the
effect of mobility on the performances of routing pro-
tocols, the definition of mobility models for simula-
tion, the definition of metrics to measure the mobility
of a given simulated network. Here, our aim is, for
any node of an operating network, to be able to de-
termine both the mobility of any other node and the
general mobility of the network.
This, of course, should be done using information
held by each node concerning the others ; namely the
routing table.
We can generally define mobility as the behav-
ior of an object (entity, person, thing, etc) that has a
changing position over time. However, in our case, it
is important to point out that, what we refer to as mo-
bility is in fact relative mobility. If two nodes move
in the same direction and with equal speed, they can
both be considered stable with respect to one another.
However, if the other agents in the network are static
then the two nodes have very high mobility relatively
to the rest of the network. Thus, two measures of
mobility may be considered: individual mobility of
nodes and global mobility of the network. In our case,
global mobility of the network should be considered
because it gives an idea on the mobility of the sur-
rounding environment. This general measure, how-
ever, does not reflect the individual mobility of each
node because it is an average measure and thus sta-
ble nodes can not be discovered using it. Thus, in
our opinion, it is also useful to consider the individual
mobility of nodes. In such a case, it will be easier for
a client agent, given a set of servers with different in-
dividual mobility measure, to choose the one it judges
most appropriate to respond to its needs. We propose
to use the same metrics to measure both general net-
work and individual node mobilities. In fact, we will
consider that the global mobility of the environment
is the average of the mobilities of all reachable nodes.
Before, examining the information that may re-
flect the mobility of a node or of the network as a
whole, it should be noted that, since we do not con-
sider that the nodes are equiped with a globally ac-
cessible location service (GPS for instance), a node is
only able to partially capture its relative mobility with
respect to other nodes and the mobility of the path be-
tween it and all reachable destination nodes. More-
over, the disappearence (respectively reappearence)
of a node from the routing table can be the conse-
quence of its relative mobility. Even though, such a
situation may also be the result of the switching off
or on of the device, we choose to consider that this
information is a hint concerning this particular node’s
mobility .
To measure mobility of nodes, we choose to con-
sider the following information gathered from the
routing table: the appearances and disappearances of
nodes in the routing table, the number of hops to all
reachable nodes and the first hop to each destination.
If we consider the situations depicted in figure 2, in
which S
represents the position of server S at time
(i [0, 3] and t
< t
), we notice that the move-
ment of S between t
and t
can be read in Cs routing
table as a change in the distance to S; the movement
between t
and t
is read by C as a change in the first
hop of the route to S; and movement between t
is seen as the disappearance of S from Cs routing
Yet, this capture is only partial as the available in-
formation in the routing table (the existence of the
destination node, the hop count and the first hop to
destination) is not sufficient to detect all mobile nodes
Figure 2: Detected mobility.
Figure 3: Undetectable mobility.
or routes. In fact, any change in a route, past the first
hop, will be unoticeable if it does not modify the hop
count. Figure 3 in which S
, S
and S
represent ,re-
spectively, the position of server S at times t
, t
< t
< t
) represents such a situation.
From Cs point of view S is always 4 hops away
and always has the same first hop to destination.It
is thus not considered mobile. For these reasons, it
would be useful, when studying mobility, to know
all intermediate nodes between C and S. This infor-
mation is available from the routing table if C and S
are neighbours (in this case there are no intermediate
nodes between C and S).Otherwise, the exact route
can only be discovered by examining the routing ta-
bles on the intermediate nodes.
To estimate the mobility of the nodes in the net-
work, we propose to consider the routing table at reg-
ular time intervals and, for each reachable destination
node, to measure the following :
Path Stability: The percentage of time during
which each entity has been in the routing table
since the begining of the time interval,
Link Stability: The percentage of time during
which each entry has been as a neighbour in the
routing table since the begining of the time inter-
Distance Stability: The average number change in
hop count to each entry since the begining of the
time interval,
First Hop Stability: The average number of
changes in the first hop since the begining of the
time interval.
4.3 Density
Density in ad hoc networks is another information
that may be considered when choosing an appropri-
ate server with which to cooperate. On the one hand,
ICEIS 2008 - International Conference on Enterprise Information Systems
if the network is dense, there will be more possibili-
ties to find the best route to each destination. On the
other hand, in a scarce network, less potential routes
will be available for each destination. The movement
of the nodes may have, as an effect, to break links in
a route. In such a case, the redundancy of routes in
a dense network will often allow to find a new route
to reach the destination node. On the contrary, such a
movement might result, in interruption of all commu-
nications with the destination if node density is not as
Mobility and density are thus closely related to
one another, and it is clear that some configurations
are more favorable than others. For instance, it seems
clear that a dense and static network will be more effi-
cient in finding stable routes to destination nodes than
a scarce and very mobile network. However, the case
of stable and scarce networks and of mobile and dense
networks need to be studied with attention.
As it is the case for mobility, different granulari-
ties can be given to the measure of density: the global
density of the network, composed of all reachable
nodes, can be considered or density can be calculated
on a per hop basis.
We are now studying the effects of both mobility
and density on the efficiency of communications in a
manet in order to determine precisely how these con-
textual attributes should be measured and how they
can be considered in trust computation. This will be
the subject of a future paper.
4.4 Server and Client Profiles
Number of hops, mobility and density are contextual
attributes that are part of the routing context; they
influence the quality of the communication between
client and server. In the classification described in
section 4, we have chosen to consider the routing con-
text as a part of the environment context. We will now
discuss how the profiles of both client and server may
also be considered as contextual information and in-
cluded in what is defined by (Schilit et al., 1994) as
the user context.
In a distributed system, nodes may be organised
in virtual communities (Abdul-Rahman and Hailes,
2000) and prefer to cooperate with nodes in the same
community or in partner communities. Our trust
model defines two types of communities : static com-
munities that reflect an underlying administrative or-
ganisation (we could for instance consider a ”stu-
dent” community, an ”administrative” community, a
”teacher” community and a ”research community
in a university’s network) and dynamic communities
which each node defines based on the results of the
interactions it has had with the others and thus on the
trust it has in them in providing a given service with
a certain level of quality. Concerning dynamic com-
munities, they can be used to regroup well behaving
servers together or, contrarily, misbehaving servers ;
for example, a node could dynamically define, both,
the community of servers in which it has ”very high”
trust for providing service S and the community of
servers in which it has ”very low” trust in providing
service S. The trust to put on an interaction with
a member of the community is defined by the dy-
namic community itself. Any node, whether server or
client, may thus belong to several communities, and
we consider that such information should be consid-
ered as profile information that may be useful in pre-
dicting the result of an interaction. In the case of static
communities a trust policy can be defined to describe
inter-community trust relation. The static communi-
ties to which belong the server and the client may also
be considered as contextual information since they
may affect the quality with which the service is pro-
4.5 First Hop
The contextual information we have considered up to
now in the routing context affect the quality of an in-
teraction, however cooperative the nodes of the net-
work are. In fact, mobility, density and the number of
hops to a destination are factors that affect the quality
of communication regardless of the willingness of the
nodes to properly route traffic. Thus, the contextual
attributes presented above do not take into account the
presence of selfish or malicious nodes in the network.
Such nodes may, by refusing to relay data, have a dis-
astrous effect on communication.
Rather than considering routing as any other ser-
vice and using the model to compute the trust in
routers, we propose to also use situational informa-
tion to take into account such malicious nodes.
Whenever two agents wish to communicate with
one another, unless they are neighbours, they must
rely on intermediate nodes and especially on a first
hop. This information may seem very partial in the
case of long routes. Yet, we believe that it is an im-
portant contextual attribute to be added to the routing
In this section we have presented contextual at-
tributes that seem interesting to consider when com-
puting trust. These contextual arguments can be or-
ganised according to the following taxonomy repre-
sented in figure 4. We have also showed in this sec-
tion that certain context attributes (hop count, mobil-
ity and density) helped dealing with the general be-
Figure 4: Taxonomy of context.
havior of manets when only in the presence of coop-
erative nodes, other attributes (profile informations)
took advantage of a possible underlying organisation
and others stills (hop count) may help in avoiding non
cooperative or malicious agents.
Contextual information should be considered
both, when recording experiences and during the trust
computing process. An experience should thus carry
the contextual information that reflects the context in
which it has occurred. Moreover, to evaluate the trust
we have in a server S that, in a specific context C,
we will receive its response to a request, we should
transpose the set of all previous experiences with S
into context C. Through this transposition, we evalu-
ate, for each experience, the quality it would have had
if it had occurred in the current context. The set of
transposed experiences can now be used as an input
to the trust computing process, to calculate the trust
assigned to S in context C.
Trust is a concept that must be studied with attention
in distributed environments based on the exchange
of services between partners. In this paper we have
briefly presented some of the models proposed by
research for computing trust. These models do not
explicitely take contextuals information into account
when assessing trust. We have discussed that in the
particular case of manets, the context, and particu-
larly the routing context, should be considered in the
trust computation process. We have identified a set
of contextual attributes that may affect the result of
an interaction with a service provider. We propose to
use these contextual attibutes together with the other
parameters that characterise each experience when as-
signing trust to a cooperation partner. These attributes
may greatly influence the way in which a client agent
perceives the quality of a given server. It may thus
be more efficient to request a service form a provider
with low trust in a favourable context than to one with
higher trust in a disadvantegeous context. Yet, before
considering the proposed context variables, an ade-
quate metric, for measuring their values should be de-
fined. Even though the metrics used to measure some
of the proposed variables seem quite straightforward
(this is the case for the number of hops between both
partners), other factors are much more tricky to mea-
sure and the definition of a metric for these variables
is not clear (this is the case for mobility). Moreover,
while some variables may be considered individually,
others are so closely related to each others that they
should be considered together (this is the case for den-
sity and mobility). We are now studying the influ-
ence of the identified context variables on the qual-
ity of communications in order to define metrics for
these attributes. A second step will be to study how
these contextual attributes should be considered by a
context-aware trust model for manets.
Abdul-Rahman, A. and Hailes, S. (2000). Supporting trust
in virtual communities. In HICSS ’00: Proceedings
of the 33rd Hawaii International Conference on Sys-
tem Sciences-Volume 6, page 6007, Washington, DC,
USA. IEEE Computer Society.
Abowd, G. D., Dey, A. K., Brown, P. J., Davies, N., Smith,
M., and Steggles, P. (1999). Towards a better under-
standing of context and context-awareness. In HUC
’99: Proceedings of the 1st international symposium
on Handheld and Ubiquitous Computing, pages 304–
307, London, UK. Springer-Verlag.
Boleng, J., Navidi, W., and Camp, T. (June 2002.). Metrics
to enable adaptive protocols for mobile ad hoc net-
works. In in Proceedings of the International Con-
ference on Wireless Networks (ICWN ’02), Las Vegas,
Nev, USA,, pages 293–298.
Castelfranchi, C. and Falcone, R. (1998). Principles of
trust for mas: Cognitive anatomy, social importance,
and quantification. In Proceeding of the International
Conference on Multi-Agent Systems(ICMAS’98).
Duma, C., Shahmehri, N., and Caronni., G. (August 2005.).
Dynamic trust metrics for peer-to-peer systems. In In
Proc. of 2nd IEEE Workshop on P2P Data Manage-
ment, Security and Trust.
Gambetta, D. (1988). Can We Trust Trust? Basil Blackwell.
Ghassemian, M., Friderikos, V., and Aghvami, A. H.
(September 2005). On mobility metrics applied for
ad hoc network protocol evaluation. In The 7th
IFIP International Conference on Mobile and Wire-
less Communications Networks (MWCN), Marrakech,
ICEIS 2008 - International Conference on Enterprise Information Systems
Grandison, T. W. (2003). Trust management for internet
applications. PhD thesis, Departement of computing,
University of London.
Hekmat, R. and Mieghem, P. V. (October 2003). Degree
ditribution and hopcount in wireless ad hoc networks.
In Proceedings of the 11th IEEE International Confer-
ence on Networks (ICON 2003), pages 603–609, Syd-
ney, Australia.
Marsh, S. P. (1994). Formalising Trust as a Computational
Concept. PhD thesis, University of Stirling.
McKnight, D. H. and Chervany., N. L. (1996). The mean-
ings of trust. Technical report, Carlson School of
Management,University of Minnesota,United States.
Nguyen, C. T. and Camp, O. (2007). A bayesian network
based trust model for improving collaboration in mo-
bile ad hoc networks. International IEEE Conference
on Computer Sciences - RIVF(07).
Salber, D., Dey, A. K., and Abowd, G. D. (1999). The
context toolkit: aiding the development of context-
enabled applications. In CHI ’99: Proceedings of the
SIGCHI conference on Human factors in computing
systems, pages 434–441, New York, NY, USA. ACM
Schilit, B., Adams, N., and Want, R. (December 1994).
Context-aware computing applications. In Proceed-
ings of IEEE Workshop on Mobile Computing Systems
and Applications.
Toivonen, S., Lenzini, G., and Uusitalo, I. (2006). Context-
aware trust evaluation functions for dynamic reconfig-
urable systems. In Proceedings of the WWW’06 Work-
shop on Models of Trust for the Web (MTW’06), Edin-
burgh, Scotland, UK.
Yu, B. and Singh, M. P. (2001). Towards a probabilistic
model of distributed reputation management. In Pro-
ceedings of the Fourth Workshop on Deception, Fraud
and Trust in Agent Societies, Montreal, Canada, pp.