Domain-specific Trust for Context-aware BDI Agents
Preliminary Work
Arthur Casals
1
, Eduardo Ferm
´
e
2
and Anarosa A. F. Brand
˜
ao
1
1
Laborat
´
orio de T
´
ecnicas Inteligentes, EP/USP, Av. Prof. Luciano Gualberto, 158 - trav. 3, S
˜
ao Paulo, SP, Brazil
2
Universidade da Madeira, Campus Universit
´
ario da Penteada, 9020-105, Funchal, Portugal
Keywords:
Context-aware Systems, Multiagent Systems, BDI, Trust, Contextual Planning, Experience Sharing, Learning.
Abstract:
Context-aware systems are capable of perceiving the physical environment where they are deployed and adapt
their behavior accordingly. Multiagent systems based on the BDI architecture can be used to process contextual
information in the form of beliefs. Contextual information can be divided and structured in the form of
information domains. Information and experience sharing enables a single agent to receive data on different
information domains from another agent. In this scenario, establishing a trust model between agents can take
into account the relative perceptions each agent has of the others, as well as different trust degrees for different
information domains. The objective of this work is to adapt an epistemic model to be used by agents with their
belief revision in order to establish a mechanism of domain-specific relative trust attribution. Such mechanism
will allow for each agent to possess different trust degrees associated with other agents regarding different
information domains.
1 INTRODUCTION
Context-aware systems are capable of capturing in-
formation from the environment and using it to adapt
their functions accordingly (Hong et al., 2009). The
information captured from the environment itself can
be referred to as context (Abowd et al., 1999), and it
can be represented and used in different ways – de-
pending on which aspects are relevant to the context-
aware system using it (Kim and Chung, 2014; Nalepa
and Bobek, 2014). Intelligent agent architectures are
among the ones that can be used by context-aware
systems (Kwon and Sadeh, 2004). The belief-desire-
intention (BDI) agent architecture (Rao and Georgeff,
1991) is of particular interest due to its inherent use
of environmental information (context) in the form
of beliefs. Beliefs are used to determine what the
agent has chosen to do (its desires), and how com-
mitted it is to that choice (its intentions) (Cohen and
Levesque, 1990). Systems composed of multiple in-
telligent agents are called Multiagent Systems (MAS)
(Wooldridge, 2009). The intelligent agents that com-
pose a MAS may interact among themselves, collabo-
rating and exchanging information in order to achieve
their objectives.
The information exchanged between agents can be
originated from the environment in which each agent
is situated, or it can be a representation of experi-
ences derived from each agent’s actions over time.
Nevertheless, the information exchanged and used by
intelligent agents is aggregated to any information
that each agent already possesses. In particular, BDI
agents use this information as part of their belief revi-
sion process - which, in a general sense, refers to the
process of altering beliefs to take into account new
acquired information (G
¨
ardenfors, 2003). When this
new information is conveyed by other agents, ques-
tions related to trust and expertise may arise: environ-
mental observations usually can be trusted (acquired
by direct observation), but other agents’ experiences
may be associated with individual trust or reputation
metrics (Huynh et al., 2006). At the same time, repu-
tation and trust are subject to pertinence: while a high
degree of trust may be associated with a given agent,
the experiences it possess can be more or less relevant
depending on the subject they refer to.
From an agent’s perspective, experience from
other agents may be more or less relevant. Since
agents in different MAS can use different subsets of
all contextual information available, their experience
relevance is also bound to the information they used
in their reasoning process. Therefore, trust and reli-
ability on experience exchange between agents may
also be associated with which subsets - or domains -
244
Casals, A., Fermé, E. and Brandão, A.
Domain-specific Trust for Context-aware BDI Agents - Preliminary Work.
DOI: 10.5220/0006600602440249
In Proceedings of the 10th International Conference on Agents and Artificial Intelligence (ICAART 2018) - Volume 1, pages 244-249
ISBN: 978-989-758-275-2
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

of the existing information are actually used by each
agent. This situation is similar to what we observe
in the real world: information provided by doctors
on medical matters and sports are subject to differ-
ent credibility degrees. While medical matters fall
within the domain of expertise of doctors, sports have
no correlation to their expertise domain. Therefore,
health diagnostics provided by a doctor are highly re-
liable, but opinions related to sports may not receive
the same degree of reliability (since the subject is not
pertinent to a doctor’s experience). We will use this
example along the text to illustrate different concepts
and aspects related to the present work.
The objective of this work is to extend an exist-
ing epistemic model to be used by context-aware BDI
agents in conjunction with their belief revision pro-
cess. This would allow for a single agent to possess
different trust degrees associated with other agents re-
garding different information domains, in a manner
that different credibility parameters can be attributed
to received experiences. In that manner, a mechanism
of trust attribution can be used in conjunction with
experience processing and multiple information do-
mains. Trust models can then be incorporated into
the agent’s planning process through the attribution of
different trust indicators to experiences received from
the same agent.
This paper is organized as follows: Section 2 de-
tails the general concepts used in this work. An exist-
ing epistemic model for multi-source belief revision is
presented in Section 3, along with the modifications
made to accommodate information domains. Dis-
cussions about the proposed model and related work
are presented in Section 4. In Section 5 we present
our considerations on the proposed model and future
work.
2 GENERAL CONCEPTS
Since the trust model is intended to be used in con-
junction with context-aware BDI agents, it is impor-
tant to understand a few concepts involving context
and information domains. The process of physically
gathering contextual information is not part of the
scope of this work.
2.1 Contextual Information
Contextual information can be collected and dis-
tributed differently. It can also be detailed and or-
ganized in different levels, depending on its intended
use. Different constraints can also determine its dis-
tribution model, such as interoperability with a pre-
existent communication model or bandwidth limita-
tions. Contextual information can be organized in dif-
ferent ways, depending on its purpose. This organi-
zation can also differ across different information di-
mensions. Mobile devices, for example, use different
sensors to gather data mostly related to physical en-
vironment aspects, such as localization and accelera-
tion. Information on the social information dimension
is limited or non-existent. On the other hand, identi-
fication cards can retain organizational data – such as
role in the company, unique identification record, and
clearance level. In this case, the social information
dimension is more detailed than in the previous one,
while the physical information dimension is almost
non-existent.
As a term, ”information domain” is broadly used
to refer to different aspects and purposes of informa-
tion organization (Hjørland, 2002). Generally speak-
ing, information domains can be used to represent
deterministic sets of information that are different
among themselves in both content and organization
(Hennessy, 1991). In the example previously pre-
sented, ”soccer” and ”medicine” are examples of two
different information domains. Depending on how the
information is organized, the content - or what is be-
ing represented - can be determined by its own repre-
sentation. An ontology, for example, can be defined
as a set of terms of interest in an information domain,
along with the relationships among these terms (Mena
et al., 1998).
Therefore, we consider that contextual informa-
tion can be composed of different information do-
mains. Sensor data gathered and aggregated by an
internal sensor network, for example, can be consid-
ered as an information domain within a given envi-
ronment. Another information domain could be rep-
resented by user preferences stored and organized in
a mobile device. When the user is in the environment,
the contextual information is composed by both infor-
mation domains - which can be used by an agent in its
reasoning process (Figure 1).
Information domains can be also used to structure
information exchanged between two or more agents -
such as past experiences. Different experiences can
contain information related to different information
domains. When we consider the information domains
”soccer” and ”medicine”, for example, past experi-
ences can describe how effective a player can be ac-
cording to the weather (soccer), or perceived phys-
ical symptoms that may lead to a specific progno-
sis (medicine). Using information domains, however,
does not change the fact that one single agent can re-
ceive conflicting information from different agents. In
that case, the agent receiving the conflicting informa-
Domain-specific Trust for Context-aware BDI Agents - Preliminary Work
245

Figure 1: Example of different information domains within
the contextual information.
tion must take into account factors such as how it can
reach a consistent conclusion, or if all involved agents
can be equally trusted, for example. This process is
called belief revision, and it is explained in more de-
tail in the next paragraphs.
2.2 Belief Revision in MAS
Belief revision is the process of changing beliefs to
take into account a new piece of information. This
is a non-trivial process, since there are several dif-
ferent ways to revise current knowledge - taking into
account a different number of factors, and following
different set of rules. The logical formalization of
belief revision is discussed in different fields of re-
search, including philosophy, databases, and artificial
intelligence (mostly for the design of rational agents)
(G
¨
ardenfors, 2003).
According to G
¨
ardenfors (G
¨
ardenfors, 2003),
there are three main methodological questions to be
settled when trying to solve belief revisions in a log-
ical manner: (i) the representation of beliefs; (ii) the
relationship between the represented beliefs (explicit)
and any other beliefs that can be derived from those
(implicit); and (iii) the logic behind discarding and
retaining existing beliefs in the revision process. In
order to solve these questions, there are a number of
integrity constraints in place that will not be detailed
in the present work. It is important, however, to high-
light how existing knowledge can be represented in
order to establish a logical belief revision process.
When an epistemic model is used to represent an
agent’s beliefs in a given moment, the epistemic state
is an idealization of the psychological concept and
represents the cognitive state of the agent in a given
moment (G
¨
ardenfors, 1988). The belief revision pro-
cess involves the study of how the agent’s knowledge
base is changed in the presence of new information.
Since it must be performed in a computational en-
vironment, it is necessary to establish a belief repre-
sentation which can be processed in conjunction with
logical operators. For that purpose, the AGM model
(Alchourr
´
on et al., 1985) is used to represent epis-
temic states in the form of belief sets, while each be-
lief is represented in the form of a single sentence.
This model predicts three different changes that can
be made to the knowledge base: (i) expansion (addi-
tion of new information), (ii) contraction (removal of
existing information), and (iii) revision (addition of
new information while preserving consistency of the
knowledge base).
In order to formally represent the beliefs, we will
use a language ζ that assumes a finite set of atomic
propositions closed under truth-functional operations.
Each element of ζ is a sentence denoted by lower-
case Greek letters. Arbitrary tautologies and contra-
dictions are represented by > an ⊥, respectively. A
consequence operator Cn is used, taking sets of sen-
tences to sets of sentences. This operator satisfies the
Tarskian properties of inclusion, monotony, and iter-
ation. Cn is also compact and satisfies the deduc-
tion theorem and supraclassicality. These properties
can be referred to as AGM-assumptions (Ribeiro and
Wassermann, 2009).
Following this formalism, the change operations
recognized by the AGM model can be described as
follows:
• Expansion: a sentence α is added to the belief set
φ. This operation is represented by φ + α;
• Contraction: a sentence α is removed from the be-
lief set φ. This operation is represented by φ - α;
• Revision: a sentence α is added to the belief set φ,
while other sentences are removed from the same
belief set in order to preserve its consistency. This
operation is represented by φ * α.
When a MAS is considered, the belief revision
process can be used either to maintain the con-
sistency of a single agent’s epistemic state or to
achieve collective goals (involving multiple agents) -
which implies maintaining the consistency of a shared
knowledge base. The first process is referred as
Multi-Source Belief Revision (MSBR), and the second
one is called Multi-Agent Belief Revision (MABR)
(Tamargo, 2012). The MSBR process - which in-
volves processing multiple sources of information
from a single agent’s perspective - will be used as a
basis for the present work.
3 EXTENDED EPISTEMIC
MODEL FOR MSBR
As previously mentioned, we will use an existing
epistemic model for MSBR as a basis for the present
ICAART 2018 - 10th International Conference on Agents and Artificial Intelligence
246

work (Tamargo, 2012). This model was developed
specifically to represent knowledge in a MAS, and
its formalism was intended to be used in conjunction
with a MSBR mechanism. Therefore, there are al-
ready formal support for belief-based operations (ex-
pansion, contraction, and revision). The revision pro-
cess also uses a credibility order to deal with even-
tual inconsistencies. Multiple existing agents are
considered as information sources (informants), and
each different agent’s knowledge base if represented
through the use of belief bases with additional (meta)
information. The formalism details present in the
original epistemic model will be explained in the next
paragraphs, along with the extensions made to the
original model in order for it to accommodate the for-
mal use of information domains.
When interacting among themselves, each agent
incorporate the other agents’ knowledge through the
use of information objects. Each information object
associates a sentence with an agent. All agents are
identified in a finite set of agents, denoted as A = {A
1
,
A
2
,..., A
n
}. Also, in order to consider different in-
formation domains in its belief revision process, it is
necessary that this information is properly formalized
and included in the original epistemic model. For that
purpose, we define the following formalism:
• Information Domain: An information domain is a
tuple D = (G, M), where G is a grammar that de-
scribes the information structure contained in the
domain and M represents all metadata associated
with the domain. While neither the grammar nor
the metadata need to be detailed for the purpose
of this work, it is important to recognize them
since they are associated with an unique identifier
to each information domain.
• Context: A context D = {D
1
, D
2
,...,D
n
} is defined
as a finite set of different information domains that
compose the contextual information, where each
information domain D
i
(1 5 i 5 n) is unique.
With these considerations in mind, the original
epistemic model definitions were extended to con-
sider information domains in its structure. Using the
formalism described above, we could consider three
agents - an engineer (A
E
), a doctor (A
D
) and a soccer
player (A
P
) - talking about different subjects. Each
subject refers to a specific information domain, in-
cluding ”soccer” (D
S
) and ”medicine” (D
M
). When-
ever the conversation goes towards each of these sub-
jects, it is expected that the engineer gives more or
less credibility to either of the other people. If we
also take into consideration the credibility labels de-
fined by ”not credible”, ”plausible”, and ”very credi-
ble” (following a strict total order C
o
), opinions from
the doctor on medicine topics tend to be perceived by
”very credible” by the engineer, while opinions on the
same subject from the soccer player tend to be ”plau-
sible” at most. This situation represents a credibil-
ity order that can be represented by A
S
<
(A
E
,D
M
)
C
o
A
M
and A
M
<
(A
E
,D
S
)
C
o
A
S
(considering that the doctor will
never be as credible as the soccer player in soccer-
related subjects, and the opposite in medicine-related
subjects).
• Information Object: An information object is a
tuple I = (α,A
i
, D
j
), where α is a sentence of a
propositional language ζ, A
i
∈ A, and D
j
∈ D. In-
formation objects are used to represent an agent’s
belief base, and can be used to associate a given
sentence to a specific agent. This allows for the
identification of the source of each information
received by an agent. The extension proposed al-
lows for information objects to associate a given
sentence to a specific agent and a specific domain
within a context D.
• Belief Base: A belief base of a given agent A
i
(1 5
i 5 n) is a set K
A
i
= {I
1
, I
2
,...,I
q
} that contains in-
formation objects (α,A
p
, D
j
) (1 5 p 5 q) received
from other agents (p 6= i) and proper beliefs (p =
i) regarding different domains in the context D.
Thus, the set κ = 2
ζ×A×D
represents all the belief
bases for all information domains within the con-
text. As an example, we can consider the finite
set of agents given by A = A
1
, A
2
, A
3
, A
4
and the
belief base for agent A
1
: K
A
1
= (β,A
1
, D
2
),(α,A
2
,
D
3
),(α,A
3
, D
1
).
• Sentence Function: The sentence function Sen (
Sen : K → (2
ζ
,D) ) over a belief base K ∈ κ
is defined as Sen(K) = {α : (α,A
i
,D
j
) ∈ K}.
For a given agent A
i
, its belief base is consis-
tent for a given domain if C
n
(Sen(K
A
)) is consis-
tent for the same domain. Considering the ex-
ample in the belief base definition, Sen(K
A
1
) =
(α,D
1
),(α,D
3
),(β,D
2
).
• Agent Identifier Function: An agent identifier
function Ag ( Ag : κ → (2
A
, D) establishes a
relationship between a belief set and a finite set of
agents for different information domains within
a context D, allowing for the identification of
agents that are referenced within a given belief
set K ∈ κ. This function is defined by: Ag(K) =
{A
i
: (α,A
i
, D
j
) ∈ K}. Considering the example
in the belief base definition,Ag(K
A
1
) = (A
1
, D
2
),
(A
2
, D
3
), (A
3
, D
1
).
• Assessment: In order to represent the credibility
that one agent assigns to other agents for every do-
main in D, an assessment function is used. Cred-
ibility, as a value assigned to an specific agent,
Domain-specific Trust for Context-aware BDI Agents - Preliminary Work
247

can be represented by a finite set of credibility val-
ues (or labels) C = c
1
,...,c
k
common to all agents.
The credibility values follow a strict total order.
Therefore, for a finite set of agents A and a cred-
ibility set C, an assessment c
A
i
,D
j
for an agent A
i
regarding a domain D
j
is a function c
A
i
: (A,D) →
C that assigns a credibility value from C to each
agent A
j
∈ A, regarding each domain D
j
in D.
Since the credibility set is common to all agents
within A, each agent possess comparable credibil-
ity values assigned by other agents. On the other
hand, since the credibility values are conditioned
to specific information domains, credibility values
regarding different information domains for the
same given agent cannot be compared. Similarly
as in the original model, different credibility val-
ues can be assigned to the same agent by different
other agents for the same information domain. For
example: considering A = A
1
, A
2
, A
3
, D = D
1
, D
2
,
and C = c
1
, c
2
, c
3
, the credibility values assigned
to A
3
regarding the domain D
1
can be different:
c
A
1
(A
3
, D
1
) = c
2
and c
A
2
(A
3
, D
1
) = c
3
. At the same
time, the credibility values for the same agent re-
garding different information domains can also be
different: c
A
1
(A
2
, D
1
) = c
2
and c
A
1
(A
2
, D
2
) = c
3
.
• Credibility Order among Agents: Since different
credibility values (following a strict total order)
can be assigned by a single agent A
i
for the same
domain D
j
, a credibility order for the other agents
can be established. In the previous example, con-
sidering A
1
and the credibility values over A
2
and A
3
regarding the information domain D
1
re-
spectively given by c
A
1
(A
2
, D
1
) and c
A
1
(A
3
, D
1
).
If c
A
1
(A
2
, D
1
) ≺ c
A
1
(A
3
, D
1
) or c
A
1
(A
2
, D
1
) =
c
A
1
(A
3
, D
1
), it means that - according to A
1
- A
3
is at least as credible as A
2
regarding the informa-
tion domain D
1
. This relationship is represented
by A
j
≤
(A
i
,D
p
)
C
o
A
k
, meaning that - according to
A
i
- A
k
is at least as credible as A
j
regarding an
information domain D
p
. Similarly, the strict re-
lationship A
j
<
(A
i
,D
p
)
C
o
A
k
can be defined, mean-
ing that A
k
is strictly more credible than A
j
re-
garding the information domain D
p
. Likewise, A
j
=
(A
i
,D
p
)
C
o
A
k
means that A
k
is as credible as A
j
for
the information domain D
p
. Therefore, the rela-
tionship ≤
(A
i
,D
p
)
C
o
is a total order over A. It is im-
portant to notice, however, that credibility order
among agents is confined to specific information
domains, and that the credibility order between
two agents can change for different information
domains.
4 DISCUSSION AND RELATED
WORK
The present work aims at serving as an initial step
towards solving the problem of the use of different
trust degrees associated with different information do-
mains in belief revision processes. While the origi-
nal epistemic model proposes a formalism that can be
used in multi-source belief revision processes, it does
not take into account the existence of different infor-
mation domains. Therefore, an agent can only receive
a global credibility (or trustworthiness) degree from
any other agent.
Also, from a learning perspective, a global credi-
bility degree can have a negative impact in the overall
belief revision process. If a given agent possess ex-
tremely accurate experiences regarding a specific ac-
tivity but no consistent or useful information about
anything else, it can be perceived as ”unreliable” by
other agents in the system. While this perception
would help in the case of the unreliable information
transmitted to the other agents, it would diminish or
even void the benefits that processing the accurate in-
formation could bring to the revised beliefs.
There are different publications on formalisms re-
lated to both multi-agent and multi-source belief revi-
sions ( (Kfir-Dahav and Tennenholtz, 1996; Liu and
Williams, 2001; Cantwell, 1998; Dragoni and Puliti,
1994)). Our decision to extend an existing belief re-
vision formalism was based on the fact that the epis-
temic model used as a reference also took into con-
sideration aspects of trust and reputation of agents in
a distributed environment. Aspects such as plausibil-
ity, reputation maintenance and information retrans-
mission in a multiagent system were addressed. Since
our long-term goal is to study these same aspects un-
der a multi-expertise perspective, it was appropriate
to adapt an existing formalism and revisit the already
studied aspects in conjunction with the new research.
As previously mentioned, our studies are also
related to existing work on agent argumentation
schemes associated with expertise ( (Melo et al.,
2016)), as well as trust and reliability aspects on MAS
( (Wang and Singh, 2007; Tamargo, 2012)). In this
work, we focus on the formalism of relative associa-
tion of trust (credibility) regarding different informa-
tion domains. Our intent is to explore the concept of
both relative and localized trust, and how it can im-
pact the experience and information sharing process
in a multiagent system.
It is also important to notice that as a preliminary
work, there are more complex questions and chal-
lenges that are not addressed yet. Different intelligent
agents in different MAS can possess different primi-
ICAART 2018 - 10th International Conference on Agents and Artificial Intelligence
248

tives, for example, which would require mechanisms
in place for experience or meaning interpretation -
which is not part of the scope of the present study.
Alternative methods for belief revision, argumenta-
tion mechanisms, or credibility evaluation processes
are also not part of the scope of the present work.
5 CONCLUSIONS AND FUTURE
WORK
The contribution of this paper resides on the extension
of an existing epistemic model in order to allow its use
by context-aware BDI agents in their belief revision
process. Using an epistemic model in conjunction
with the concept of information domains provides the
formalization necessary for a multi-source belief re-
vision process based on contextual information. The
use of such model allows for a single agent to possess
different trust degrees associated with other agents re-
garding different information domains.
While more complex problems are not addressed
in this work, we intend to use the extended epis-
temic model presented here as a basis for future re-
search. This will include further development of the
extended epistemic model and the implementation of
a MSBR mechanism to be used by context-aware
agents, along with trust calculation and conflict solv-
ing mechanisms that can benefit from this model.
ACKNOWLEDGEMENTS
Eduardo Ferm
´
e is partially supported by FCT
MCTES and NOVA LINCS UID/CEC/04516/2013,
FCT SFRH/BSAB/127790/2016 and FAPESP
2016/13354-3. Arthur Casals is supported by CNPq,
grant no. 142126/2017-9.
REFERENCES
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 Interna-
tional Symposium on Handheld and Ubiquitous Com-
puting, pages 304–307. Springer Berlin Heidelberg.
Alchourr
´
on, C. E., G
¨
ardenfors, P., and Makinson, D.
(1985). On the logic of theory change: Partial meet
contraction and revision functions. The journal of
symbolic logic, 50(2):510–530.
Cantwell, J. (1998). Resolving conflicting informa-
tion. Journal of Logic, Language and Information,
7(2):191–220.
Cohen, P. R. and Levesque, H. J. (1990). Intention is
Choice with Commitment. Artificial Intelligence,
42(2–3):213–261.
Dragoni, A. F. and Puliti, P. (1994). Distributed belief re-
vision versus distributed truth maintenance. In Tools
with Artificial Intelligence, 1994. Proceedings., Sixth
International Conference on, pages 499–505. IEEE.
G
¨
ardenfors, P. (1988). Knowledge in flux: Modeling the
dynamics of epistemic states. The MIT press.
G
¨
ardenfors, P. (2003). Belief revision, volume 29. Cam-
bridge University Press.
Hennessy, P. (1991). Information domains in CSCW. Stud-
ies in Computer Supported Cooperative Work: The-
ory, Practice and Design, Eds. JM Bowers and SD
Benford, Elsevier.
Hjørland, B. (2002). Domain analysis in information sci-
ence: eleven approaches–traditional as well as inno-
vative. Journal of documentation, 58(4):422–462.
Hong, J.-y., Suh, E.-h., and Kim, S.-J. (2009). Context-
aware systems: A literature review and classification.
Expert Systems with Applications, 36(4):8509–8522.
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.
Kfir-Dahav, N. E. and Tennenholtz, M. (1996). Multi-agent
belief revision. In Proceedings of the 6th conference
on Theoretical aspects of rationality and knowledge,
pages 175–194. Morgan Kaufmann Publishers Inc.
Kim, J. and Chung, K.-Y. (2014). Ontology-based health-
care context information model to implement ubiqui-
tous environment. Multimedia Tools and Applications,
71(2):873–888.
Kwon, O. B. and Sadeh, N. (2004). Applying case-
based reasoning and multi-agent intelligent system to
context-aware comparative shopping. Decision Sup-
port Systems, 37(2):199–213.
Liu, W. and Williams, M.-A. (2001). A framework for
multi-agent belief revision. Studia Logica, 67(2):291–
312.
Melo, V. S., Panisson, A. R., and Bordini, R. H. (2016).
Trust on beliefs: Source, time and expertise. In
TRUST@ AAMAS, pages 31–42.
Mena, E., Kashyap, V., Illarramendi, A., and Sheth, A.
(1998). Domain specific ontologies for semantic in-
formation brokering on the global information infras-
tructure. In Formal Ontology in Information Systems,
volume 46, pages 269–283. Amsterdam: IOS Press,
MCB UP Ltd.
Nalepa, G. J. and Bobek, S. (2014). Rule-based solution
for context-aware reasoning on mobile devices. Com-
puter Science and Information Systems, 11(1):171–
193.
Rao, A. S. and Georgeff, M. P. (1991). Modeling rational
agents within a BDI-architecture. In Allen, J., Fikes,
R., and Sandewall, E., editors, Proceedings of the 2nd
International Conference on Principles of Knowledge
Representation and Reasoning, pages 473–484. Mor-
gan Kaufmann publishers Inc.: San Mateo, CA, USA.
Ribeiro, M. M. and Wassermann, R. (2009). Agm revision
in description logics. Proceedings of ARCOE.
Tamargo, L. H. (2012). Knowledge dynamics in multi-agent
systems: Plausibility, belief revision and forwarding
information. AI Communications, 25(4):391–393.
Wang, Y. and Singh, M. P. (2007). Formal trust model for
multiagent systems. In IJCAI, volume 7, pages 1551–
1556.
Wooldridge, M. (2009). An introduction to multiagent sys-
tems. John Wiley & Sons.
Domain-specific Trust for Context-aware BDI Agents - Preliminary Work
249