Presentations of the


AAAI Spring Symposium on

Agent-Mediated Knowledge Management (AMKM-03)

Stanford University

March 24-26, 2003, Stanford University

This page assembles abstracts, slides and pictures from presentations given at AMKM-03.
These and a couple of additional papers can be found in the
AAAI Technical Report SS-03-01.
Revised  and extended versions will be published as Post-Proceedings in the
Springer LNAI Series (LNAI 2926).AMKM Frontpage
For a short report to appear in the AI Magazine, click here.
Interested in doing AMKM research or applications? Click here ;-)

Monday 24

Title + Link2Abstract
Ludger van Elst, Virginia Dignum
1 2 3
Charles J. Petrie
Invited Talk: Adding AI to Web Services
1 2
M. Bonifacio, P. Bouquet, G. Mameli, M. Nori Peer-mediated Distributed Knowledge Management mameli.ppt
1 2
A. Graesser, M. Ventura, G. T. Jackson, J. Mueller, X. Hu,
N. Person
The Impact of Conversational Navigational Guides on the Learning, Use, and Perceptions of Users of a Web Site AAAI 2003 Jackson.ppt
R. S. S. Guizzardi, L. Aroyo, G. Wagner
Agent-oriented Knowledge Management in Learning Environments: A Peer-to-Peer Helpdesk Case Study
S. Schulz, R. Kalcklösch, T. Schwotzer, K. Herrmann
Towards Trust-based Knowledge Management for Mobile Communities
N. Santo, I. Haimowitz AMKM for Internet Behavior Tracking aaai-ss-2003-Ira-nuno-pfizer.ppt
1 2
E. Moreale, S. Watt An Agent-Based Approach to Mailing List Knowledge Management
M. S. Ackerman,K. DeMaagd, S. Cotterilla, A. Swensona I-DIAG: From Community Discussion to Knowledge Distillation


Coffee Break

1 2 3 4 5

Breakout Sessions

1 2 3 4 5 6 7

Tuesday 25

Author/Presenter Title + Link2Abstract Slides Pictures
J. M. Dodero, S. Arroyo, V. R. Benjamins Dynamic Generation of Agent Communities from Distributed Production and Content-driven Delivery of Knowledge
1 2
W.F. Lawless Information Density Functional Theory (IDFT): A quantum approach to knowledge fusion and organizational mergers lawless_aaai_s03_amkm.ppt
J. J. Vasco Furtado, V. Ponte Machado Improving Organizational Memory through Agents for Knowledge Discovery in Database Furtado.ppt
K. Hui, S. Chalmers, P. Gray, A. Preece Experience in using RDF in Agent-mediated Knowledge Architectures
J. A. Pinto de Magalhães, C. J. Pereira de Lucena Using an Agent-Based Framework and Separation of Concerns for the Generation of Document Classification Tools Presentation_Magalhaes.ppt
J. Louçã Modeling context-aware distributed knowledge Louca_SanFrancisco2.pdf
J. Novak, M. Wurst, M. Schneider, M. Fleischmann, W. Strauss Discovering, visualizing and sharing knowledge through personalized learning knowledge maps – an agent-based approach
D. A. Evans,G. Grefenstette, Y. Qu, J. Shanahan, V.  Sheftel Agentized, Contextualized Filters for Information Management
AAAI-03 ViviDocs Final Slides.pdf

Breakout Sessions

1 2 3 4 5 6 7 8 9

Coffee Break

T.-D. Cao, F. Gandon Integrating external sources in a corporate semantic web managed by a multi-agent system AMKM_2003_V02.ppt
W. Krueger, J. Nilsson, T. Oates, T. Finin Automatically Generated DAML Markup for Semistructured Documents
1 2
M. Ehrig, C. Schmitz, S. Staab, J. Tane, C. Tempich Towards Evaluation of Peer-to-Peer-based Distributed Knowledge Management Systems AMKM2003_Tempich_AIFB_Karlsruhe.ppt
K. Rawlins, C. Eliot, V. Lesser, B. Woolf, Peter Cassin Ontology extraction for educational knowledge bases
S. Toivonen, H. Helin Representing Interaction Protocols in DAML
A. Pease
AMKM at the Plenary Session
Plenary session.ppt
1 2 3

Informal Meeting at the Gordon Biersch, Palo Alto

1 2 3 4

Wednesday 26

Author/Presenter Title + Link2Abstract Slides Pictures
A. Molani, A. Perini, E. Yu, P. Bresciani Analyzing the Requirements for Knowledge Management using Intentional Analysis
S. C. Bailin,W. Truszkowski Perspectives: An Analysis of Multiple Viewpoints in Agent-Based Systems
Sorry, pic file broken
C. A. Tacla, J.-P. Barthès A Multi-Agent Architecture for Knowledge Acquisition
A. Pease, J. Li Agent-Mediated Knowledge Engineering Collaboration

Final Presentations of Breakout Groups
AMKM Standardization.ppt

F. Gandon, A. Pease, J.-P. Barthès, D. A. Evans
"Fake" Panel Discussion
(Please, do not take contents too seriously!)
Semantic-Web Final Slides.pdf
1 2

You can download all photos in better resolution as one big (>80MB) ZIP file:
Sorry that some of the pictures are quite dark. A gamma correction in a photo editor may help.


Peer-mediated Distributed Knowledge Management

Distributed Knowledge Management is an approach to knowledge management based on the principle that the multiplicity (and heterogeneity) of perspectives within complex organizations is not be viewed as an obstacle to knowledge exploitation, but rather as an opportunity that can foster innovation and creativity. Despite a wide agreement on this principle, most current KM systems are based on the idea that all perspectival aspects of knowledge should be eliminated in favor of an objective and general representation of knowledge. In this paper we propose a peer-to-peer architecture (called KEx), which embodies the principle above in a quite straightforward way: (i) each peer (called a K-peer) provides all the services needed to create and organize “local” knowledge from an individual’s or a group’s perspective, and (ii) social structures and protocols of meaning negotiation are introduced to achieve semantic coordination among autonomous peers (e.g., when searching documents from other K-peers). A first version of the system, called KEx, is implemented as a knowledge exchange level on top of JXTA.

The Impact of Conversational Navigational Guides on the Learning, Use, and Perceptions of Users of a Web Site

Knowledge management systems will presumably benefit from intelligent interfaces, including those with animated conversational agents. One of the functions of an animated
conversational agent is to serve as a navigational guide that nudges the user how to use the interface in a productive way. This is a different function from delivering the content of the material. We conducted a study on college students who used a web facility in one of four navigational guide conditions: Full Guide (speech and face), Voice Guide, Print Guide, and No Guide. The web site was the Human Use Regulatory Affairs Advisor (HURAA), a webbased facility that provides help and training on research ethics, based on documents and regulations in United States Federal agencies. The college students used HURAA to complete a number of learning modules and document
retrieval tasks. There was no significant facilitation of any of the guides on several measures of learning and performance, compared with the No Guide condition. This
result suggests that the potential benefits of conversational guides are not ubiquitous, but they may save time and increase learning under specific conditions that are yet to be isolated.

Agent-oriented Knowledge Management in Learning Environments: A Peer-to-Peer Helpdesk Case Study

In this paper we present an analysis and modeling case study for agent mediated knowledge management in educational environments: Help&Learn, an agent-based peer-to-peer helpdesk system to support extra-class interactions among students and teachers. Help&Learn expands the student’s possibility of solving problems, getting involved in a cooperative learning experience that transcends the limits of classrooms. To model Help&Learn, we have used Agent-Object-Relationship Modeling Language (AORML), an UML extension for agent-oriented
modeling. The aim of this research is two-fold. On the one hand, we aim at exploring Help&Learn’s potential to support collaborative learning, discussing its knowledge management strategy. On the other hand, we aim at showing the expressive power and the modeling strengths of AORML.

Towards Trust-based Knowledge Management for Mobile Communities

Agent-mediated Knowledge Management is a promising approach to handle and maintain knowledge, especially in a distributed and mobile environment. A natural example for such an environment is a community: individuals that group together because they share the same interests. Trust is an important issue when building communities on sharing knowledge.
This paper presents our conceptual framework for trustbased knowledge management with respect to the highly distributed and mobile environment of communities. We extend an existing knowledge management system by using mobile agents that serve as dynamic delegates of the mobile user in different locations. These agents may exchange knowledge with other agents. When the user returns to a location a previously delegated agent resides, this agent reconciles the newly found knowledge with the user’s knowledge base. Similar to sharing knowledge, trust plays a central role in this process since it controls the autonomous knowledge acquisition of a user’s agents.

AMKM for Internet Behavior Tracking

We have developed a real-world business application performing agent-mediated knowledge management (AMKM) for the purposes of tracking Internet behaviors.  Our application touches aspects of the following application areas: Analysis and design methods for AMKM, agents for group formation and awareness, agent-based workflow in the KM context, organizational implications of agent use in KM.  We believe that this system illustrates several successful examples of agent-mediated KM in tracking and analyzing a complex set of drivers impacting Internet sites for the pharmaceutical industry and the application of Agents in the formalization of knowledge as key input for marketing business intelligence. 

An Agent-Based Approach to Mailing List Knowledge Management

The widespread use of computers and of the internet have brought about human information overload, particularly in the areas of internet searches and email management. This has made Knowledge Management a necessity, particularly in a business context. Agent technology – with its metaphor of agent as assistant – has shown promise in the area of information overload and is therefore a good candidate for Knowledge Management solutions. This paper illustrates a mailing list Knowledge Management tool that is centred around the concept of a mailing list assistant. We envisage this system as the first step towards a comprehensive agentbased Knowledge Management solution.

I-DIAG: From Community Discussion to Knowledge Distillation

I-DIAG is an attempt to understand how to take the collective discussions of a large group of people and distill the messages and documents into more succinct, durable knowledge. I-DIAG is a distributed environment that includes two separate applications, CyberForum and Consolidate. The goals of the project, the architecture of IDIAG, and the two applications are described here.

Dynamic Generation of Agent Communities from Distributed Production and Content-driven Delivery of Knowledge

In an agent-mediated distributed knowledge management system, knowledge is firstly produced and then delivered to a person or community of users that is interested in it. A common issue to this aim is the setting-up of communities that can drive the delivery of contents. In our study, knowledge-producing agents are arranged into separate interaction domains and a distributed interaction protocol is used to consolidate knowledge that is generated therein. Knowledge produced in this way can be used as the source data to dynamically build user communities that drive the delivery of knowledge amongst users. The multi-agent architecture presented here can be applied to the shared creation of various electronic resources, like learning objects, e-books or software artifacts.

Information Density Functional Theory (IDFT): A quantum approach to knowledge fusion and organizational mergers

Agent mediated knowledge management (AMKM) for multiple agent systems must address the generation of information, I, decision-making, the derivation of knowledge, K, and their relationship with agent organizations, requiring that trust, cooperation, and competition also be addressed. With the discovery of a group process rate equation, to satisfy these requirements an ab initio approach is used for organization formation based on first principles and then linked with K fusion
and organizational mergers.

Improving Organizational Memory through Agents for Knowledge Discovery in Database

In this article we describe a computational architecture called MC2 that brings together a set of tools that contribute to the knowledge management process by allowing for the creation and maintenance of an organizational memory. To achieve this end, the approach taken by MC2 is aimed at establishing favorable conditions for interaction between the personnel within an organization as well as with the system itself. In particular, we describe agents that attempt to achieve automatic knowledge discovery from organizational databases and from the manner in which that knowledge is integrated into the MC2 environment. By using these agents, organizational
development is fostered through the dissemination of knowledge, yet, in such a way so as to be transparent to the holders of that knowledge without requiring additional activities above and beyond those already carried out as part of their day-to-day routine.

Experience in using RDF in Agent-mediated Knowledge Architectures

We report on experience with using RDF to provide a rich content language for use with FIPA agent toolkits, and on RDFS as a metadata language. We emphasise their utility for programmers working in agent applications and their value in Agent-Oriented Software En gineering. Agent applications covered include Intelligent Information Agents, and agents forming Virtual Organisations. We believe our experience vindicates more direct use of RDF, including use of RDF triples, in programming knowledge architectures for a variety of applications.

Using an Agent-Based Framework and Separation of Concerns for the Generation of Document Classification Tools

The present paper describes the Avestruz1 framework, a multi-agent based architecture for generating tools for document classification in specific domains. We have designed and implemented an architecture that separates the concerns related to document search and election (called a concern platform) from those related to the classification algorithm. Not only is it possible to use and experiment with existing classification algorithms but also to generate new algorithms, taking into consideration specific characteristics of the domain.

Modeling context-aware distributed knowledge

This paper presents a multi-agent model to support decisionmaking in organizations. The model is characterized by being interactive, distributed, and incremental and by the use of cognitive maps to represent the knowledge of decision-making actors. The main proposition is to consider the context of concepts belonging to cognitive maps in a way that it represents agent’s mental states, allowing some kind of inference. To do so, context is conceptualized in cognitive maps, defining agent’s mental states from concepts being causally related to their context.

Discovering, visualizing and sharing knowledge through personalized learning knowledge maps – an agent-based approach

This paper presents an agent-based approach to semantic exploration and knowledge discovery in large information spaces by means of capturing, visualizing and making usable implicit knowledge structures of a group of users. The focus is on the developed conceptual model and system for creation and collaborative use of personalized learning knowledge maps. We use the paradigm of agents on the one hand as model for our approach, on the other hand it serves as a basis for an efficient implementation of the system. We present an unobtrusive model for profiling personalised user agents based on two dimensional semantic maps that provide 1) a medium of implicit communication between human users and the agents, 2) form of visual representation of resulting knowledge structures. Concerning the issues of implementation we present an agent architecture, consisting of two sets of asynchronously operating agents, which enables both sophisticated processing, as well as short respond times necessary for enabling interactive use in real-time.

Agentized, Contextualized Filters for Information Management

Every time a user engaged in work reads or writes, the user spontaneously generates new information needs: to understand the text he or she is reading or to supply more substance to the arguments he or she is creating. Simultaneously, each Information Object (IO) (i.e., word, entity, term, concept, phrase, proposition, sentence, paragraph, section, document, collection, etc.) encountered or produced creates context for the other IOs in the same discourse. We present a conceptual model of Agentized, Contextualized Filters (ACFs)—agents that identify an appropriate context for an information object and then actively fetch and filter relevant information concerning the information object in other information sources the user has access to. We illustrate the use of ACFs in a prototype knowledge management system called ViviDocs.

Integrating external sources in a corporate semantic web managed by a multi-agent system

We first describe a multi-agent system managing a corporate memory in the form of a corporate semantic web. We then focus on a newly introduced society of agents in charge of wrapping external HTML documents that are relevant to the activities of the organization, by extracting semantic Web annotations using tailored XSLT templates.

Automatically Generated DAML Markup for Semistructured Documents

The semantic web is becoming a realizable technology due to the efforts of researchers to develop semantic markup languages such as the DARPA Agent Markup Language (DAML). A major problem that faces the semantic web community is that most information sources on the web today lack semantic markup. To fully realize
the potential of the semantic web, we must find a way to automatically upgrade information sources with semantic markup. We have developed a system based on the STALKER algorithm that automatically generates DAML markup for a set of documents based on previously seen labeled training documents. Our ultimate goal is to
develop a program that performs DAML markup for arbitrary documents given a particular ontology.

Towards Evaluation of Peer-to-Peer-based Distributed Knowledge Management Systems

Distributed knowledge management systems (DKMS) have been suggested to meet the requirements of today’s knowledge management. Peer-to-peer systems offer technical foundations for such distributed systems. To estimate the value of P2P-based knowledge management evaluation criteria that measure the performance of such DKMS are required. We suggest a concise framework for evaluation of such systems within different usage scenarios. Our approach is based on standard measures from the information retrieval and the databases community. These measures serve as input to a general evaluation function which is used to measure the efficiency of P2P-based KM systems. We describe test scenarios as well as the simulation software and data sets that can be used for that purpose.

Ontology extraction for educational knowledge bases

A student who wishes to learn about some particular topic does not have many options. An often used tool is the search engine, which gives a tiny and difficult to control window into the vast amounts of information that is available on the internet. A student who wants to learn some concept should be able to interact with the available information in a coherent and personalized way. The classroom is the ideal of this goal, and our system would not replace, but augment it. It is within the reach of modern tutoring systems to use both knowledge of the student and of the subject’s structure in order to present a subject in a manner that is more coherent
and pedagogically sound than currently existing technology. One of the basic building blocks of such a system is the model of topic structure, and most importantly, how to obtain the information that fills the model. Here we outline our research platform for the study of ontology lifecycle management, as well as several techniques that have so far had qualitative success. This research is taking place within the context of the Digital Libraries Initiative, under which thousands of instructional objects are
organized, ranging from multimedia tutors , to lecture notes and papers. Our long term goal is to develop agent based tutoring systems which draw on this large knowledge base, and we have discussed our approach to this in other recent work.

Representing Interaction Protocols in DAML

We present an extension to DAML-S for representing interaction protocols. An interaction protocol defines the messaging patterns between communicating entities such as software agents. Serializing interaction protocols in a suitable form for reuse supports creating software agents capable of adapting to various environments. Serialized interaction protocols can be utilized, for example, when specifying details of interaction between a contractor and a subcontractor operating in the Internet.

Analyzing the Requirements for Knowledge Management using Intentional Analysis

Novel approaches to knowledge management aim at exploiting knowledge properties such as its distributed and local nature, and to consider organizations as set of actors cooperating and competing to pursue private as well as common goals. Therefore, designing knowledge management solutions requires a deep analysis of the interests and intents of strategic organizational actors, and of the dependency relationships among them. This paper describes an approach based on intentional modelling techniques. Two examples from a hospital case study are used to illustrate.

Perspectives: An Analysis of Multiple Viewpoints in Agent-Based Systems

This paper summarizes our ongoing effort to understand and characterize the uses of multiple perspectives in space science. We have attempted to bring together a variety of theories and research areas practiced by a variety of communities with apparently very limited cross-fertilization to date. For example, the combination of different levels of reasoning is widely recognized as being desirable in sensor data fusion applications (Hall & Llinas, 2001). However, we have not found a unifying account of the variety of possible approaches. In this effort, we try to perform such an accounting using the notion of ontology. Ontologies serve as a unifying device in a study of perspectives because they help to formalize the notion of an interpretation. There have been previous formalizations of the notions of observation and interpretation, such as Bennett et al (1989), but these have focused mostly on the quantitative treatment of uncertainty, rather than on the logical structure of an interpretation. Ontologies help in providing a logical treatment. By doing so, they provide a framework for integrating higher-level reasoning approaches.

A Multi-Agent Architecture for Knowledge Acquisition

This paper concerns a multi-agent system for knowledge management (KM) in research and development (R&D) projects. R&D teams have no time to organize project information, or to articulate the rationale behind the actions that generated the information. Our aim is to provide a system for helping team members to make knowledge explicit, and to allow them to share their experiences, i.e., lessons learned (LL), without asking them too much extrawork. The article focuses on how we intend to help the team members to feed the system with LL, using the day-to-day operations they perform on desktop computers, and how we intend to exploit the LL by using a case-based reasoning engine.

Agent-Mediated Knowledge Engineering Collaboration

Knowledge Management is most necessary and valuable in a collaborative and distributed environment. A problem with commercial knowledge management tools is that they do not understand at a deep level the content that they are managing. In this paper we discuss the System for Collaborative Open Ontology Production (SCOOP), which manipulates logic expressions and checks for redundancies or contradictions between the products developed by different engineers. SCOOP also includes an automated workflow process that supports recommendations for changes and voting to agree on changes.

Draft of short report to appear in AI Magazine

Has Knowledge Management research given enough attention and importance to Agent technology? Have agents researchers considered the potentialities and demands of the KM field as an application domain? Such questions are increasingly being asked and several projects have been started which attempt to provide an answer.
The aim of the Agent-Mediated Knowledge Management symposium was therefore to bring together researchers and practitioners of both fields to discuss benefits, possibilities and added-value of a cross-fertilization between KM and Agent technology.
Knowledge Management already is an important topic in business studies for more than a decade. From the starting days of KM, technology has been recognized as an enabling, and often even a leading, factor for connecting (e.g., people to other people or knowledge) and converting (e.g., data into knowledge). Comprehensive KM endeavors, however, have always realized that KM is primarily a management science, and not a computer science. This implies a different role for technology in KM, that of supporting and extending human interaction and learning, and therefore a need for intelligence-enhanced, integrated and personalized solutions.
Which, as any agent researcher will tell, are exactly the aims and characteristics of agents…
The link is therefore established and the starting point for AMKM participants to discuss and present their own research launched.
The symposium set off with a keynote talk by Charles J. Petrie from Stanford University who illuminated the relationship between AI and Web Service technology - assets and challenges as well as drawbacks. In the symposium, twenty-five talks were grouped into six  presentation sessions:
During these sessions ongoing research, finished projects and position papers from industry and academics were presented to a lively and inquisitive audience which provided nice interactions and debates within and outside the conference room. Furthermore, Stanford’s outside campus facilities (together with the gorgeous Californian weather) provided an excellent surrounding for breakout sessions during which participants discussed comprehensive themes in small groups. Topics for these sessions were the semantic web, standardization questions, tools and methods for AMKM, and further research directions. Again here, lively discussions and interaction were prevalent, and some of the groups are considering the production of articles describing their discussion and conclusions.
As the workshop evolved  it became increasingly clear that even though the main premises and objectives were shared by all, different participants held different views on the field and on the interaction between KM and agent technology. Of course, one could a priori expect to be able to classify some participants more as ‘agent-people’, that see agents as the ultimate solution for KM, and while others are more ‘KM-people’, for which agents may be an interesting possibility for KM. However, another – in AI well-known – polarity soon became apparent: that between the ‘statistical-people’ and the ‘cognitive-people’. The symposium therefore ended with a hilarious panel discussion for which a member of each of these four fields was asked to adopt and defend the exactly opposite view to the one he usually would take.
In summary, the overall consensus was that cross-fertilization between KM and agent technology is a theme to be further developed as the possibilities for research and application ahead are countless. A follow-up workshop is certainly to be considered.

Ludger van Elst (DFKI Kaiserslautern), Virginia Dignum (Achmea and University of Utrecht), Andreas Abecker (DFKI Kaiserslautern).

Interested in doing AMKM research or applications?

The steps may be steep...

The steps may be steep...

..., but the door is open! (And we can already see the light ;-)

So, don't hesitate contacting us.

For comments, suggestions, questions
Last update of this page: April 15, 2003