You are on page 1of 7

ESMAP: A Multi-Agent Platform for Extending a Knowledge

Management System
Graciela Garca, Ruth Cobos
Escuela Politcnica Superior, Universidad Autnoma de Madrid, 28049, Madrid, Spain.
Graciela.Garcia@estudiante.uam.es, Ruth.Cobos@uam.es

Abstract experience, and expertise to empower people to make the


most appropriate decisions.
For more than one decade Knowledge Management As the problem of information overload has become
has taken advantage of some abilities that software agents more acute, a need for more powerful and flexible devices
are endowed with. Agent features such as autonomy, has arisen. It is then that, software agents begin to be
cooperation, communication, and learning capacity have incorporated to Knowledge Management applications;
been used to improve the performance of Knowledge they can be considered as an alternative and a successful
Management applications. manner to tackle the information issue. Agents are used to
In this paper we outline our proposal for search networks to select and alert users of new relevant
implementing a Jade-based multi-agent platform to information. As well, they are able to filter out less
enhance the potential of KnowCat: a fully consolidated, relevant information from information inputs, or to
thoroughly tested and validated Knowledge Management produce recommendations to users, among other functions
system which has been in active use at Universidad [26], [27].
Autnoma de Madrid (Spain) since 1998. We also give a Agent-based technologies for Knowledge Management
succinct description of the agent platform architecture applications are not novel. These systems have developed
and a brief presentation of its current status. and evolved through the last decade, and because of their
suitability to manage the information quagmire, have
1. Introduction gained public and widespread acceptance. Some examples
of relevant proposals are described in [18], [17], [16], [20]
Information has become the most available product to and [1]. We briefly discuss these approaches in section 2.
web users nowadays. Hence, the amount of information Our working scenario is the KnowCat system.
that people are faced to handle, is growing much faster KnowCat is a fully consolidated and thoroughly tested
than their capacity to process it [23]. Because of that, the and validated Knowledge Management system which has
working place has turned out to be as entropic as never been in active use at Universidad Autnoma de Madrid
before. The continuous stream of incoming information to (Spain) since 1998. Different experiments have been
the workplace is not a new phenomenon. The relatively carried out to test and validate the KnowCat system. The
new issue is the speed with which information arrives and results of these experiments are extensively described in
the rate with which information changes. Information [2], [10], [11], [12] and have demonstrated that KnowCat
technology has made a big contribution to this situation is a successful approach.
by providing tools such as Internet, Intranets, email, etc. These auspicious findings have encouraged our line of
Too often users are overwhelmed by too much work and thus motivated the use of agents in KnowCat.
information without being able to determine how relevant We believe that an agent-based approach can provide
it can be and how much impact it may have on their daily some extra benefits to the system in terms of added value
work. to users. Especially due to the fact that knowledge is
In order to overcome this quagmire, it is essential to distributed among Knowcat knowledge nodes. Because of
implement procedures and technologies to ensure the that, the search and analysis of knowledge by manual
provision of relevant customised information to every means can be quite laborious and so more than simple
user [16]. Knowledge Management is expected to supply human intervention may be required. Agents can relieve
the necessary support for understanding and users from this kind of tasks
communicating the appropriate information that users The organisation of this paper is as follows. In the next
may need to perform their tasks. In this framework, section we review some related work. In Section 3 we
Knowledge Management support entails the complete present KnowCat the knowledge management system
information life cycle: creation, integration, sharing and upon which we build our proposal. In Section 4 we
distribution. It must be kept in mind that Knowledge discuss the proposed approach: ESMAP Platform. In
Management does not suppose a mere implementing Section 5 we briefly describe the main enabling
technology but a way of consciously using information, technology used to implement our prototype. In Section 6
we make a brief presentation of the ESMAP architecture.
In section 7 we outline the Expert Recommender Module.

Proceedings of the 2006 IEEE/WIC/ACM International Conference


on Web Intelligence (WI 2006 Main Conference Proceedings)(WI'06)
0-7695-2747-7/06 $20.00 2006
In section 8 we deal with the current status of our architecture supports interaction between distributed,
proposal. Finally, in Section 9 we present our conclusions heterogeneous agents and is built on top of the Voyager
and we offer a perspective of our future work in this area. platform which is extended towards an agent platform by
the addition of directory and broker services,
2. Related Work administration tools and agent classes.
Preece and colleagues propose the fusion of knowledge
A relatively vast amount of investigation has been from multiple, distributed and heterogeneous sources in
carried out in the field of agent-based Knowledge KRAFT (Knowledge Reuse And Fusion/Transformation))
Management (KM) applications. We have mentioned [20]. KRAFT has an agentbased architecture, in which
before some of the approaches that we consider as all knowledge processing components are software
milestones in this area. In this section we briefly review agents. The architecture uses constraints as a common
some of these systems. knowledge interchange format, expressed in a common
One of the earliest approaches to use agents in KM ontology. Knowledge contained in local sources is
applications was Liebermans proposal: Letizia [18]. translated into the common constraint language, fused
Letizia, is a single-agent system that operates with a with knowledge from other sources, and then used to
conventional Web browser. The agent tracks the user's solve a specific problem, or to deliver some information
browsing behaviour (follow links, initiate searches, or to users. The generic framework of KRAFT architecture
request for help) and tries to predict what items may be of can be utilised over a wide range of knowledge domains.
interest to the user. Letizia uses a simple set of heuristics For example, it has been used in a network data services
to model the user's browsing behaviour. When requested, application as well as in prototype systems for advising
Letizia may provide a page containing its students on university transfers, and for advising health
recommendations, which the user may choose either to care practitioners on drug therapies. The implementation
follow or disregard. As we see, Letizia is basically a of KRAFT is based on the FIPA standard with RDF as a
document recommender. content language.
Next in time, Konstan et al. [17] propose a A more recent proposal is the one presented by
homogeneous multiagent system called GroupLens. Abecker et al. in [1]. This approach called the Frodo
Homogeneous indicates that the system has mostly one project postulates the concept of Organisational Memory.
type of agents; although agents may not have the same An Organisational Memory can be viewed as a meta-
goals, their tasks and capabilities are quite similar. This information system with multiple ontology-based
approach consists of a system for collaborative filtering of structures and a workflow-based context representation
netnews. Its main goal is to help people find articles that [1].
will interest them among a big mass of available articles. Each Organisational Memory (OM) is structured in a
Users rate articles after reading them. GroupLens uses four-layer architecture:
these ratings for two purposes. First, to correlate the The application layer manages the process context in
ratings in order to determine which users' ratings are most form of weaklystructured workflows.
similar to each other. Second, to predict how well users The source layer contains information sources with
will like new articles based on ratings from similar users. different degrees of formalisation (processes models,
The construction of user profiles is implied. The core of text documents, etc.)
GroupLens is an open client-server architecture in which The knowledge description layer provides uniform
news clients are used to enter ratings and to display access to the sources by means of ontologies.
predictions; in turn rating servers are used to distribute The knowledge access layer connects the application
ratings and to deliver predictions. Basically, GroupLens with the source layer.
can be considered as a collaborative filtering As a means to cope with complex and heterogeneous
recommender. environments and at the same time to keep control of the
Koch and colleagues have provided a later approach system behaviour, FRODO agents are organised in
with an outstanding degree of refinement considering the societies. Consequently, FRODO agents are described in
time in which the proposal was presented. The aims of the terms of its knowledge, goals and competencies, as well
Campiello project described in [16] are: to improve the as in terms of its rights and obligations. FRODO is
connection among local inhabitants of historical places, to implemented in JADE [15].
prompt these people to have an active role in the creation Our approach is mainly inspired by the theoretical
of cultural information and to support new and better work of Bonifacio et al. [5] where a distributed
connections with cultural managers and tourists. intelligence paradigm (DIP) for knowledge management
Campiello provides a recommender module, a search is proposed. Our main contribution to this direction of
module, and a shared data space. Agent architecture is research is an implementation of DIP adapted to the
designed to ensure an easy integration, adaptation and Knowcat environment. In addition, our proposal supports
extensibility of these components. The Campiello a relatively wider range of services.

Proceedings of the 2006 IEEE/WIC/ACM International Conference


on Web Intelligence (WI 2006 Main Conference Proceedings)(WI'06)
0-7695-2747-7/06 $20.00 2006
3. KnowCat: The Working Scenario work. ESMAP is a cooperative multi-agent platform
designed to run in a distributed environment and is
KnowCat (Knowledge Catalyser) is a distributed non- expected to enable the provision of the following new
supervised system that supports collaborative Knowledge services:
Management for knowledge communities. KnowCat is in Automatic mapping of documents and subsequent
use at Universidad Autnoma de Madrid at the present classification in the KnowCat knowledge tree.
time, and according to the researchers that have developed Documents are processed using content-based
it, its main goal is to empower the crystallisation of analysis techniques and as a result feature vectors
collective knowledge [2], [10], [11]. are obtained. Next, vectors are compared to certain
KnowCat provides a highly structured and organised stored patterns and then, according to the best
framework to find relevant and good quality knowledge matching, documents are classified and placed into
that may be of interest for knowledge communities. the best fitting knowledge subtree [3], [4], [25].
KnowCat comprises a set of topic-specific knowledge Dynamic information exchange between KnowCat
centres called KnowCat knowledge nodes. Each KnowCat knowledge nodes: Mapping documents into feature
knowledge node can be accessed through a specific URL vectors that are then stored in a repository, allows
using a Web browser. the fast identification and location of documents
Within KnowCat, knowledge is organised in a tree related to a certain topic. A specialised agent (node
structure (the knowledge tree). The root of this tree agent) attached to each node is responsible for
corresponds to the main topic of a knowledge area which letting other node agents know about the knowledge
receives the name of KnowCat knowledge node. content of its node and eventually exchange
Consequently, each node of the knowledge tree is information about a searched topic. Furthermore,
univocally related to just one specific topic. As well, each meaning negotiation may be implied [5].
node contains two items: a set of alternative descriptions Document recommendation. Content analysis
of the topic (a set of links to Web documents) and a techniques and the construction of user profiles are
refinement of the topic (a list of other nodes or KnowCat to be used to produce document recommendation to
knowledge nodes that can be viewed as sub-topics of the users [14], [22], [24].
current topic). As we mentioned before, the main purpose Refine crystallisation values by incorporating new
of KnowCat is to enable the crystallisation of collective implicit interest indicators to the crystallisation
knowledge as the result of user interactions [11]. computation. According to [9], users behaviour can
The concept of knowledge crystallisation results from be a good source of implicit information about what
the concurrence of three factors: knowledge usage, users find interesting during their interaction with
endurance of knowledge across time and user opinions web pages. Some actions that users carry out can be
about knowledge. Initially, when a piece of knowledge is considered as positive interest indicators, while
introduced into the system, KnowCat assigns a low value others may be classified as negative interest
of acceptance to it. The degree of acceptance of a piece of indicators. Actions such as saving a page as a file,
knowledge is directly related to how much that piece of emailing a page, printing it, copying and pasting,
knowledge is appreciated by the community of users [11]. searching in the page for text, scrolling, spending
Since at the beginning a piece of knowledge is not known more time on a page, following a link or visiting the
by users, KnowCat is expected to give a low value of same page repeatedly can be regarded as positive
acceptance to it. interest indicators. Conversely, the absence of
The degree of acceptance of a piece of knowledge positive interest indicators can be judged as a sign of
increases as it becomes popular among users and as it negative interest [9]. More realistic crystallisation
receives their favourable opinions. Consequently a piece values are to be obtained by taking into account
of knowledge is considered crystallised if it is highly some of the implicit sources of interest that we
accessed by a great deal of users as time goes by, and if remark above.
user opinions about it are advantageous [11]. According Automatic extraction of social opinions supplied by
to this, the knowledge crystallisation process is actually users regarding document quality: Knowcat provides
the result of users contributions through their active a mechanism to collect user assessments or weight
interaction with KnowCat. This fact emphasises the assertions about every document located in the
collaborative aspect of the system and in that sense we repository. Users are allowed to select one of a few
can say that KnowCat is user oriented. provided options about different parts or general
Keeping users in mind and trying to provide means to characteristics of a document, such as title,
add more value to them, we propose the implementation conclusions, content, structure, etc. For each of the
of ESMAP (Extended Services Multi-Agent Platform). above-mentioned elements, users are prompted to
We think of it as a way to enhance the potential of rate different document aspects such as originality,
KnowCat and its construction is the main objective of this length, interest, correctness, etc. Values may range

Proceedings of the 2006 IEEE/WIC/ACM International Conference


on Web Intelligence (WI 2006 Main Conference Proceedings)(WI'06)
0-7695-2747-7/06 $20.00 2006
from 0 (lowest value) to 10 (the highest one). The at the stage of information processing [8]. It is a fact
main improvement to achieve in this service is to that terms used by certain groups of individuals to
automatically extract meaningful information such classify information items, may not serve as
as social opinions, from the analysis of these appropriate retrieval indication for others in different
assessments. That information can be then used, to situations or task domains. Mapping documents into
help authors increase the degree of acceptance of feature vectors is expected to minimise this kind of
their documents as well as significantly reduce the vocabulary problems and thus facilitate information
time needed to create good quality documents. exchange.
Expert recommendation: In the KnowCat Help users to access distributed knowledge, which
environment an expert is defined as a user who has may be located across the different KnowCat
supplied a piece of knowledge, which has knowledge nodes. Thanks to the information
crystallised [2], [11]. Common users may seek exchange that takes place among the Node agents,
experts as a source of information to complement or accessing remote information is completely
replace other sources such as documents or transparent for users.
databases. Expert recommendation intends to find
relevant experts by identifying relationships between 4. The Proposed Approach
experts and documents. In our proposal the basic
mechanism involves the participation of an Information spaces are complex by nature, basically
intelligent agent that keeps track of the documents due to the intrinsic characteristics of information itself.
that users visit in order to extract contacts in some This sort of environments is highly dynamic as
specific areas. Those contacts as well as their area of information is constantly changing and since the amount
expertise are later presented to users [19]. See of the new incoming information is often unpredictable
section 7. [7]. To add more complexity to the scenario, users needs
In the following section we present the problems that are also under permanent change themselves. All these
ESMAP and the built-in services that we have just components meet together in KnowCat and that is why it
detailed are expected to tackle. becomes an attractive and challenging environment for
We have stated earlier that the fore mentioned services the use of agents. Agents can assist users in dealing with
add value to users, but more specifically in what sense such a continuous flow of information.
may they assist users? What are the benefits that users The proposal is then to build a platform of adaptive
may perceive as a consequence of the proposed autonomous agents, which can be responsive to the
implementation? We believe that the most outstanding information needs that users may have. In keeping with
improvements are the following: this direction agents can automate repetitive tasks, process
Reduction of information overload: As documents complex data as means to generate valuable information,
and experts are recommended, users are prompted to learn from users, and provide recommendations to them.
go right to the point instead of having to wander In order to get their job done, agents are expected to
aimlessly among a mass of documents to find out if have roles and according to them may develop the
they are relevant to their search or not. For that following basic abilities:
reason users are relieved from the burden of dealing Interact with front-end users through user-friendly
with large amounts of data. interfaces.
More efficient use of time: Because of the twofold Acquire relevant data from distributed repositories.
aspect of recommendation (documents and experts), Perform information retrieval and data mining when
users are not required to spend a lot of time scanning necessary.
uninteresting documents or searching for appropriate Cooperate with other agents in order to accomplish
contacts when trying to find an expert. On the other collaborative tasks.
hand, authors may as well be guided throughout the Two important issues that derive from the stated above
process of document creation thanks to the feedback and that we address later in this paper are the following:
that comes from users assessments. As a How single agents should relate with each other in
consequence, time needed to plan and develop order to maximally exploit their capacities, so that
documents can be drastically decreased. the systems global goals can be accomplished
Fewer tasks to perform: The proposed system efficiently.
undertakes some of the tasks that used to be users How information and knowledge exchange should
responsibility, especially those related with take place among agents so that the system can
information search. successfully respond to users needs and to
Minimisation of vocabulary problem: Due to unexpected situations.
differences in vocabulary usage among knowledge
communities, efficiency can be dramatically reduced

Proceedings of the 2006 IEEE/WIC/ACM International Conference


on Web Intelligence (WI 2006 Main Conference Proceedings)(WI'06)
0-7695-2747-7/06 $20.00 2006
While the first issue deals greatly with communication source ontology editor for ontology definition and Jade
matters, the second one focuses on cooperation aspects Bean Generator to generate ontology classes.
but both imply the development of a clear coordination
strategy. 6. ESMAP Architecture
5. Enabling Technologies ESMAP provides a unified environment in which
different kinds of agents are integrated. These agents
Various enabling technologies are used to implement interact among them to retrieve, filter, and combine
the ESMAP prototype. The most important one is JADE information from distributed network-based information
[15]. The main reason for this selection is because of its sources and to provide suggestions to users while they
following attributes: portability, openness, high scalability interact with the system. A first version of the multi-agent
and a built-in messaging system. platform has been implemented and is structured as
Java Agent DEvelopment Framework (JADE) is an shown in Figure 1.
open source agent-based middleware, fully implemented
in Java and FIPA compliant. Jade provides a basal
framework together with a set of packages, to enable the User Agent
construction of multi-agent systems.
Agent platforms developed in Jade can be distributed
User
across different machines; they may even support Expert
different operative systems and are customisable through Agent
a remote graphical interface. As Jade complies with FIPA
specifications, it provides some required components for
Broker Agent
agent management:
DF (Directory Facilitator): provides a yellow-page
service. As so, DF registers every service that is
provided within the agent platform.
AMS (Agent Management System): provides a
Node Node Node
white-page service. Besides, it is responsible of Agent1 Agent2 Agentn
agent lifecycle and agent directory services.
ACC (Agent Communication Channel): controls
message exchange among agents. Information
Agents are created by instantiating a class, which Agent Databases
extends the Agent class. Agents are identified within the
agent platform by a unique global name. In order to be
Figure 1. ESMAP Architecture
useful, agents are expected to perform certain tasks, that
means to exhibit certain predefined behaviours. All tasks
The system has different generic types of agents: user
are instances of the Behaviours class.
agents, expert agents, node agents, information agents and
Jade agent communication is based on the
a broker agent.
asynchronous message-passing paradigm, which complies
with the FIPA ACL [13] standard. Messages are instances The broker agent acts as an intermediary for all
of the ACLMessage class. Agents construct and decode messages between any couple of agents.
messages internally. To facilitate agent communication, User agents are interface agents that interact with
JADE offers various already implemented FIPA-specified users capturing user actions and providing results to
interaction protocols; these are Query, Request and them. Also, user agents observe user
Contract-Net. behaviours/reactions and build current users profiles
It is required that agents share semantics so that according to their observations.
communication between agents can take place. Expert Agents provide expert lists to users based on
Exchanged messages must have a content, which has to the calculation of expert status of users and on user
be written in a certain language, and which must refer to profiles that user agents supply.
the same ontology. JADE provides support for using both Node agents handle the information related to
content languages and ontologies. Building ontologies KnowCat knowledge nodes.
manually can be an additional burden when designing Information agents have models of information
multi-agent systems; JADE allows the utilisation of some sources, information access strategies and
tools for assisting the automatic construction of correspondent task agents to which the information
ontologies. Some examples are Protg [21] an open must be transferred.

Proceedings of the 2006 IEEE/WIC/ACM International Conference


on Web Intelligence (WI 2006 Main Conference Proceedings)(WI'06)
0-7695-2747-7/06 $20.00 2006
Two further issues are under our consideration at the
present time. The first one is if an additional mechanism 6 5
for scheduling activities when an agent receives some User
kind of input may be necessary or not. The second one is
1
to determine a controlled way of dealing with error Expert List
conditions such as lost messages.
2
7. The Expert Recommender Module User Agent
Accessed
The main purpose of expert recommendation is to Documents
facilitate the location of expertise knowledge. 3
Consequently, this module is expected to propitiate access User 4 Expert Agent
to knowledge in all forms, including knowledge held by Profile
people (implicit knowledge) [28]. As we mentioned
before, in the KnowCat environment, any user who has
supplied a piece of knowledge that has crystallised is Figure 2. The Expert Recommender Module
considered an expert. According to this and keeping in
mind that the degree of acceptance of documents may
vary with time, the expert status of users must be 8. Current Status
reviewed on regular basis.
When a certain user logs into the system, his user We have already prototyped the largest part of the
agent is created and insofar as this user accesses different agent platform architecture as described here, together
documents his user agent builds and updates his current with the recommender module as well. At the present
user profile. In this way, user profiles are automatically time we are ready to begin the testing phase which
generated without the need to ask users to rate or fill up comprises two main areas of activities: 1) data collecting
questionnaires that are hard to maintain across time. and 2) analysis and discussion of results.
In Figure 2 we present the structure of the expert Data will be gathered from user activities on the web
recommender module, showing the main actors that take involving some active KnowCat knowledge nodes.
part in it, as well as the most relevant inputs and outputs Subsequent analysis and interpretation of the data
of the process. from these areas may lead to important actions. If
The expert recommender module consists of a main validation of the proposal is successful we will continue
agent, the expert agent, which interacts with other agents to work in this direction, otherwise substantial changes
such as user and information agents to elaborate a will addressed.
recommendation list. The expert agent is created right
after the user agent, and is responsible for calculating the 9. Conclusions and Future Work
expert status at that time and for providing the
corresponding expert list to the user. To build the expert We have presented the general framework and design
list, the expert agent must consider both: the user profile decisions regarding our multi-agent platform for the
and the expert status of users at a given moment in the extension of KnowCats current services. We believe and
system. In order to avoid redundant calculation of expert expect that our ideas make a positive contribution to
status, a control mechanism has been devised. The expert support collaborative knowledge management in the
list that is provided to users contains the following data: KnowCat environment. Just like Knowcat that was
expert names, degree of expertise of each expert, topic of successfully validated by exhaustive experimentation [2]
expertise and the corresponding email addresses. [10] [11] [12], a series of experiments will be conducted
Automatic user profile building and automatic in order to determine the suitability of the ESMAP
calculation of expert status have some advantages such as: proposal.
a) relieve users from the burden that entails providing Concerning future work and insofar as other services
rates or answering questionnaires and b) avoid operations are implemented, parallel exploration of relevant issues
of continuous user profile updating. Besides the such as agent coordination and negotiation becomes
information about user interests can be considered more critical. Furthermore, there are some research issues that
reliable since it is obtained based on real facts instead of still remain open from our point of view. For example,
on user opinions. those related with the use of agent learning. We want to
explore and experiment with different learning schemes in
order to determine the best fitting for our working
environment. Regarding this issue, our personal pending
question is to find out if the use of learning has a direct

Proceedings of the 2006 IEEE/WIC/ACM International Conference


on Web Intelligence (WI 2006 Main Conference Proceedings)(WI'06)
0-7695-2747-7/06 $20.00 2006
impact on the global performance of the KnowCat- related papers of Interaccin 2004, R. Navarro-Prieto, and J.
ESMAP association. Lors (Eds.), Springer: Dordrecht, The Netherlands, 2005, pp.
Initial results of our investigations, where we measure 281-290.
the impact of the Multiagent Platform on system [13] FIPA: Foundation for Intelligent Physical Agents.
http://www.fipa.org/. 07/2006.
performance, will be analysed and reported before the end [14] N. Good, J.B. Schafer, J.A Konstan, A. Borchers, B.M.
of the year. Depending on the results obtained through Sarwar, J.L Herlocker, and J. Riedl. Combining collaborative
data collection, analysis of results and interpretation, new filtering with personal agents for better recommendations. In
areas of future research may arise. Proceedings of the 1999 Conference of the American
Association of Artificial Intelligence (AAAI-99), 1999, pp. 439
Acknowledgements 446.
[15] JADE: Java Agent Development Framework.
http://jade.tilab.com. 07/2006.
This work has been partly funded by the Spanish [16] M. Koch. Knowledge Management and Knowledge
National Plan of R+D, project number TSI2005-08225- Agents in Campiello. In Proceedings of the Workshop on
C07-06. Intelligent Agents in CSCW, B. Lees, H. Muller, and C. Branki
(Editors), Dortmund, Germany, 1998, pp. 4452.
10. References [17] J.A Konstan, B.N. Miller, D. Maltz, J.L. Herlocker, L.R.
Gordon and J. Riedl. GroupLens: Applying Collaborative
Filtering to Usenet News Communications of ACM, vol. 40,
[1] A. Abecker, A. Bernardi, and L. van Elst. Agent technology no. 3, 1997, pp. 77-87.
for distributed organizational memories. In Proceedings of the [18] H. Lieberman. Letizia: An Agent That Assists Web
5th International Conference On Enterprise Information Browsing. In Chris S. Mellish, editor, Proceedings of the
Systems, Vol. 2, 2003, pp. 310. Fourteenth International Joint Conference on Artificial
[2] X. Alamn, and R. Cobos. KnowCat: a Web Application for Intelligence (IJCAI-95), Montreal, Quebec, Canada, August
Knowledge Organization. In: Lecture Notes in Computer 1995. Morgan Kaufmann publishers Inc.: San Mateo, CA, USA,
Science 1727, Chen et.al. (Eds), Springer Verlag: Berlin, 1995, pp. 924929.
Heidelberg, New York, 2000, pp. 348-359. [19] D. McDonald and M.S. Ackerman. Expertise
[3] N.J. Belkin and W.B. Croft. Retrieval Techniques. In: Recommender: A Flexible recommendation system and
Williams, M.E., Annual Review of Information Science and architecture. In Proceeding of the ACM Conference on
Technology 22(), Elsevier & ASIS, New York, 1987, pp. 109- Computer Supported Cooperative Work, Philadelphia, PA,
145. December 2-6th, 2000, pp. 231-240.
[4] N.J. Belkin, and W.B. Croft. "Information filtering and [20] A. Preece, K. Hui, A. Gray, P. Marti, T. Bench-Capon, D.
information retrieval", Communications of the ACM, 35, 1992, Jones, and Z. Cui. The KRAFT Architecture for Knowledge
pp. 29-37. Fusion and Transformation. Knowledge Based Systems, 13(2-
[5] M. Bonifacio, P. Bouquet and A. Manzardo. A Distributed 3), 2000, pp. 113120.
Intelligence Paradigm for Knowledge Management. AAAI [21] PROTEGE: http://protege.stanford.edu. 07/2006
spring symposium on bringing knowledge to business processes, [22] P. Resnick and H.R. Varian. Recommender Systems.
Stanford University, Palo Alto Calif., March 2000. Communications of the ACM, 40 (3), 1997, pp.56-58.
[6] U.M. Borghoff, and R. Pareschi. Information Technology [23] B. Sarwar, G. Karypis, J. Konstan and J. Riedl. Item-based
for Knowledge Management. Springer. Berlin, Heidelberg, collaborative filtering recommendation algorithms. In
New York, 1998. Proceedings of the Tenth International World Wide Web
[7] U.M. Borghoff, and J. Schlichter. Computer-supported Conference, Hong Kong, China. 2001.
Cooperative Work. Springer Verlag: Berlin, Heidelberg, New [24] J.B. Schafer, J. Konstan and J. Riedl. Recommender
York, 2000. Systems in E-Commerce. In: EC 99: Proceedings of the First
[8] H. Chen. Collaborative systems: Solving the vocabulary ACM Conference on Electronic Commerce, S. Schmitt and R.
problem. Computer, 27(5), 1994, pp. 5866. Bergmann, Denver, CO, 1999, pp. 158-166.
[9] M. Claypool, P. Le, M. Waseda and D. Brown. Implicit [25] J. Thorsten. Text Categorization with Support Vector
Interest Indicators. In Proceedings of ACM Intelligent User Machines: Learning with Many Relevant Features. In
Interfaces Conference (IUI), Santa Fe, New Mexico, January Proceedings of ECML-98, 10th European Conference on
2001, pp. 14-17. Machine Learning. 1998.
[10] R. Cobos and X. Alamn. KnowCat: a Knowledge [26] M. Wooldridge and N.R. Jennings. Intelligent Agents:
Crystallisation Tool. In: E-business: Key Issues, Applications Theory and Practice. In The Knowledge Engineering Review,
and Technologies, Stanford-Smith, B. and Kidd, P. T. (Eds). 10(2), 1995, pp. 115152.
IOS Press, 2000, pp. 374-380. [27] M. Wooldridge. Intelligent Agents. In Gerhard Weiss,
[11] R. Cobos. Mecanismos para la cristalizacin del editor, Multiagent Systems: A Modern Approach to Distributed
conocimiento, una propuesta mediante un sistema de trabajo Artificial Intelligence, The MIT Press, Cambridge, MA, USA,
colaborativo (Mechanisms for the Crystallisation of 1999, pp. 2778.
Knowledge, a proposal using a collaborative system). Ph. D. [28] D. Yimam-Seid and A. Kobsa. Expert Finding Systems for
thesis. Universidad Autnoma de Madrid. 2003. Organizations: Problem and Domain Analysis and the DEMOIR
[12] R. Cobos and M. Pifarr. Learning among Equals in the Approach. Journal of Organizational Computing and
Net: Analysis of KnowCat supporting group work. In: HCI Electronic Commerce, 13(1), 2003, pp.124.

Proceedings of the 2006 IEEE/WIC/ACM International Conference


on Web Intelligence (WI 2006 Main Conference Proceedings)(WI'06)
0-7695-2747-7/06 $20.00 2006

You might also like