USING AGENTS TO CONFRONT SOME OF THE CHALLENGES
OF KNOWLEDGE MANAGEMENT SYSTEMS
Javier Portillo-Rodríguez
1
, Aurora Vizcaíno
1
, Juan Pablo Soto
2
and Mario Piattini
1
1
ALARCOS Research Group, Information Systems and Technologies Department
UCLM-INDRA Research and Development Institute, Escuela de Informática, Universidad de Castilla-La Mancha
Paseo de la Universidad 4 - 13071 Ciudad Real, Spain
2
Mathematics Department, University of Sonora
Blvd. Luis Encinas y Rosales s/n, Col. Centro, CP. 83000, Hermosillo, México
Keywords: Agent architecture, Knowledge management systems, Communities of practice, Trust models.
Abstract: The importance of knowledge management has, in recent years, led to the incorporation of Knowledge
Management Systems (KMS) into companies. Some of these KMS could be considered as Recommender
Systems that are able to recommend knowledge, which is part of the company’s intellectual capital.
However, these KMS are not always welcome in the company, since the knowledge is not stored by using a
quality control, or because employees feel that these kinds of systems, rather then helping them, cause them
extra work. In this paper we present an agent architecture combined with a trust algorithm trying to avoid
some of the problems that appear when a KMS is introduced into companies.
1 INTRODUCTION
In recent years, knowledge has become an extremely
important factor (Hansen and Kautz, 2004). Subjects
such as Knowledge Management are, therefore,
currently of particular interest to organizations who
are concerned about their employees’ learning and
competitiveness, since a suitable management of
knowledge can help them to increase their members’
collaboration and encourage them to share
knowledge. At present organizations must operate in
a climate of rapid market change and high
information volume, which increases the necessity to
create knowledge management systems (KMS) that
support the knowledge process. It is possible to
consider certain Recommender Systems as KMS,
however, these kinds of systems are not always
welcomed by a company’s employees because
(Lawton, 2001) on occasions employees waste a
considerable amount of time searching for
information, with regard to this, sometimes there is
no quality control with regard to the KOs
(Knowledge Objects) introduced into the system and
employees may introduce information into the
systems which is not very valuable.
Our work is focused on attempting to reduce the
impact of these problems. We therefore use software
agents to search for information on behalf of users,
and these agents are in charge of recommending the
most suitable knowledge to them.
We pretend to use our proposal in Communities
of Practice (CoPs) which are a natural means of
sharing knowledge, which is considered to be a
critical factor for an organization’s competitive
advantage (Hansen and Kautz, 2004).
However, nowadays, these kind of communities,
due to globalization, are geographically distributed
and there are no face-to-face interactions. If CoP
members are distributed and they do not know the
other members trust between CoP members
decrease. This situation could be a problem because
people in general prefer to exchange knowledge with
“trustworthy people” and if there is not enough trust
among members knowledge exchange could
decrease too. People with a consistently low
reputation will eventually be isolated from the
community since others will rarely accept their
justifications or arguments and will limit their
interactions with them. This issue, plus the problems
pointed out previously, have led us to develop an
agent architecture and a recommendation algorithm
249
Portillo-Rodríguez J., Vizcaíno A., Pablo Soto J. and Piattini M. (2010).
USING AGENTS TO CONFRONT SOME OF THE CHALLENGES OF KNOWLEDGE MANAGEMENT SYSTEMS.
In Proceedings of the 2nd International Conference on Agents and Artificial Intelligence - Agents, pages 249-252
DOI: 10.5220/0002699202490252
Copyright
c
SciTePress
to encourage the reuse of knowledge in CoPs. In
order to tackle these problems, we have developed
an agent architecture and a trust algorithm with
which to rate KOs and Knowledge Sources (KSs)
that produce these KOs. The software agents will
therefore use this algorithm in order to decide
whether a KO or KS should be recommended to a
particular user.
Therefore in Section 2 the agent architecture is
described and later, in Section 3, a recommender
system and the recommender algorithm used by this
system is explained. . Finally, our conclusions are
outlined in Section 4.
2 AN AGENT ARCHITECTURE
The agent architecture proposed is composed of two
levels: reactive and deliberative-social. The reactive
level is considered by other authors to be a typical
level that an Agent Architecture must have (Ushida,
1998). A deliberative level is often also considered
as a typical level, but a social level is not often
considered in an explicit manner, despite the fact
that these systems (MAS) are composed of several
individuals, the interactions between them and the
plans constructed by them. The social level is only
considered in those systems that attempt to simulate
social behaviour. Since we wish to emulate human
feelings such as trust when working in CoPs, we
have added a social-deliberative level that considers
the social aspects of a community and which takes
into account the opinions and behaviour of each of
the members of that community.
Each of these levels is explained in greater detail
in the following sub-sections.
2.1 Reactive Level
This is the level in charge of perceiving changes in
its environment and responding to these changes at
the precise moment at which they occur, i.e., when
an agent executes another agent’s request without
any type of reasoning.
The components of the reactive level are (see
Figure 1):
Internal Model. This component stores the
individuals’ features. These features will be
consulted by other agents in order to discover more
about the person represented by the User Agent
Beliefs. This module is composed of inherited
beliefs (pre-defined beliefs) and lessons learned
(obtained by interaction with the environment) from
the agent itself.
Figure 1: Reactive Level.
Interests. These are a special kind of beliefs. This
component represents individual interests that an
agent has with regard to a topic or a knowledge
source.
Behaviour Generator. This component is
fundamental to our architecture. It is here that the
actions to be executed by the agent are triggered.
Depending on the information received from the
Interpeter module the agent makes a matching
process to select the correspondent behaviour.
2.2 Deliberative-Social Level
At this level, the agent has a type of behaviour
which is oriented towards objectives, that is, it takes
the initiative in order to plan its performance with
the purpose of attaining its goals.
The components of the deliberative-social level
are (see Figure 2):
Figure 2: Deliberative-Social Level.
ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence
250
Goals Generator. Depending on the state of the
agent, this module must decide what the most
important goal to be achieved is.
Social Beliefs. This component represents a view
that the agent has of the communities and their
members, for instance, beliefs about other agents.
Social Interests. This is a special type of belief. In
this case it represents interest in other agents.
Intuitions. We often trust more in people who have
similar features to our own. Thus, in this research,
intuition has been modelled according to the
similarity between agents’ profiles: the greater the
similarity between one agent and another, the greater
the level of trust. The agents’ profiles may change
according to the community in which they are
working. This factor will be used in those cases
when the agent doesn’t have enough information to
know if a KS is trustworthy.
Plan Generator. This component is in charge of
evaluating how a goal can be attained, and which
plans are most appropriate to achieve this.
Trust Generator. This module is in charge of
generating a trust value for the knowledge sources
with which an agent interacts in the community. To
do this, the trust generator module considers the trust
model explained in detail in (Soto et al, 2007) which
considers the information obtained from the internal
model and the agent’s intuitions.
3 A RECOMMENDER SYSTEM
A recommender system has been developed in order
to test the trust model and the multi-agent
architecture. In this system each CoP member is
represented by a software agent called a User Agent.
A new community member must first join a
community, which is done by using the “Register”
menu and choosing a community from those which
are available. Once registered, a member can
provide new KOs or use those which are already
available in the community and/or propose new
subjects. One way to obtain KOs in a community is
requesting a KO recommendation. To obtain a KO
recommendation user has to use the “Recommend”
menu and select a topic. To make the
recommendation, the prototype will use a
recommendation algorithm that has been design as
follows.
The input the algorithm is a set of KOs. Each
KO may or may not have been evaluated previously,
signifying that a KO may already have a list of
evaluations (along with the identity of each person
who evaluated it), or it may not have any
evaluations. This aspect will be taken into account
by the algorithm, which therefore distinguishes two
groups:
Group 1 (G1): This group is formed of the KOs
that have already been evaluated. This is the most
important group since if the agents have previous
evaluations of a KO they have more information
about it, which facilitates the task of discovering
whether or not its recommendation is advisable.
Group 2 (G2): these KOs have not been used
previously so the agents do not have any previous
evaluations of them. Let us now observe how each
group is processed by the algorithm.
In G1 the KOs will be ordered by a
Recommendation Rate which is calculated by the
User Agent for each KO. Hence RR
k
signifies the
Recommendation Rate for a particular KO called k,
and is obtained from:
(1)
where TE
i
is the pondered mean of the evaluations
determined by the trust that an agent “i” has in each
evaluator (the person who has previously evaluated
that KO). TE
i
is calculated as:
(2)
Therefore, TS
ij
is the trust value that the User
Agent “i” has in the knowledge source “j”, since in a
CoP the source which provides a KO will usually be
a CoP member. TS
ij
therefore represents the trust
that an agent “i” has in another agent “j” and E
jk
is
the evaluation that an agent “j” has made with regard
to a particular KO “k”.
The parameter TS
ik
used in Formula (1) similarly
indicates the trust that an agent “i” has in a
knowledge source “k”. In other words, the agent
must take two things into consideration when
calculating the RR
K
The other agents’ opinions of a KO “k”
pondered by the trust that agent “i” has in the
person who provided that evaluation.
The opinion that the agent “i” has in the agent
that has provided the KO “k”.
Both w1 and w2 are weights which are used to
adjust the formula. The sum of w1 and w2 should be
1.
Group 2 will use another formula to calculate the
RR
k
for each KO since, in this case, there are no
USING AGENTS TO CONFRONT SOME OF THE CHALLENGES OF KNOWLEDGE MANAGEMENT SYSTEMS
251
results of previous evaluations of the KOs. This
formula, not explained due to space problems,
basically uses a pondered mean of the trust values
that other agents have about the KS.
4 CONCLUSIONS
CoPs are a means of knowledge sharing. However,
the knowledge that is reused should be valuable for
its members, who might otherwise prefer to ignore
the documents that a community has at its disposal.
In order to encourage the reuse of documents in
CoPs, in this work we propose a multi-agent
recommender system with which to suggest
trustworthy documents. Some of the advantages of
our system are:
The use of agents to represent members of the
community helps members to avoid the problem
of information overload since the system gives
agents the ability to reason about the
trustworthiness of the other agents or about the
recommendation of the most suitable documents
to the members of the community. Users are
not, therefore, flooded with all the documents
that exist with regard to a particular subject, but
their agents filter them and recommend only
those which are most trustworthy (when they
have rates) or those which are provided by more
trustworthy sources or sources which have
preferences and features that are similar to those
of the user in question.
The system can detect those users with the
greatest level of participation and those whose
documents have obtained higher rates. This
information can be used for two purposes:
expert detection and/or recognition of
fraudulent members who contribute with
worthless documents. Both functionalities imply
various advantages for any kind of organization,
i.e., the former permits the identification of
employee expertise and measures the quality of
their contributions, and the latter permits the
detection of fraud when users contribute with
non-valuable information.
The system facilitates the exchange and reuse of
information, since the most suitable documents
are recommended. The tool can also be
understood as a knowledge flow enabler
(Rodríguez-Elias et al, 2007), which encourages
knowledge reuse in companies.
Furthermore, the proposed algorithm is quite
flexible since in many situations weights are used to
modify the formulas. This algorithm could,
therefore, be used by the designers of other
recommender systems who could decide what values
they should give to these weights in order to adapt
the formula to their needs.
REFERENCES
Hansen, B. and Kautz, K., (2004), "Knowledge Mapping:
A Technique for Identifying Knowledge Flows in
Software Organisations", in Software Process
Improvement, LNCS 3281, Springer: 126-137.
Lawton, G., (2001), "Knowledge Management: Ready for
Prime Time?", in Computer, vol. 34(2): 12-14.
Rodríguez-Elias, O., Martínez-García, A., Vizcaíno, A.,
Favela, J., and Piattini, M. (2007). "A Framework to
Analyze Information Systems as Knowledge Flow
Facilitators", in Information Software Technology,
vol. 50(6): 481-498.
Soto, J. P., Vizcaíno, A., Portillo, J., and Piattini, M.,
(2007), "Applying Trust, Reputation and Intuition
Aspects to Support Virtual Communities of Practice",
in 11th International Conference on Knowledge-Based
and Intelligence Information and Engineering Systems
(KES), LNCS 4693, Springer: 353-360.
Ushida, H., Hirayama, Y., and Nakajima, H., (1998),
"Emotion Model for Life like Agent and its
Evaluation", in Proceedings of the Fifteenth National
Conference on Artificial Intelligence and Tenth
Innovative Applications of Artificial Intelligence
Conference (AAAI'98 / IAAI'98), Madison,
Wisconsin, USA.
ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence
252