Cooperative information agents III, , matthias klusch, onn m shehory, gerhard weiß, 1999 2

413 0 0
  • Loading ...
1/413 trang
Tải xuống

Thông tin tài liệu

Ngày đăng: 08/05/2020, 07:00

Lecture Notes in Artificial Intelligence Subseries of Lecture Notes in Computer Science Edited by J G Carbonell and J Siekmann Lecture Notes in Computer Science Edited by G Goos, J Hartmanis and J van Leeuwen 1652 Berlin Heidelberg New York Barcelona Hong Kong London Milan Paris Singapore Tokyo Matthias Klusch Onn M Shehory Gerhard Weiss (Eds.) Cooperative Information Agents III Third International Workshop, CIA’99 Uppsala, Sweden, July 31 - August 2, 1999 Proceedings 13 Series Editors Jaime G Carbonell, Carnegie Mellon University, Pittsburgh, PA, USA Jăorg Siekmann, University of Saarland, Saarbrăucken, Germany Volume Editors Matthias Klusch DFKI German AI Research Center Ltd Stuhlsatzenhausweg 3, D-66123 Saarbrăucken, Germany E-mail: klusch@dfki.de Onn M Shehory The Robotics Institute, Carnegie Mellon University 5000 Forbes Ave., Pittsburgh, PA 15213, USA E-mail: onn@cs.cmu.edu Gerhard Weiss Institut făur Informatik, Technische Universităat Măunchen D-80290 Măunchen, Germany E-mail: weissg@in.tum.de Cataloging-in-Publication data applied for Die Deutsche Bibliothek - CIP-Einheitsaufnahme Cooperative information agents III : third international workshop ; proceedings / CIA ’99 Uppsala, Sweden, July 31 - August 2, 1999 Matthias Klusch (ed.) - Berlin ; Heidelberg ; New York ; Barcelona ; Hong Kong ; London ; Milan ; Paris ; Singapore ; Tokyo : Springer, 1999 (Lecture notes in computer science ; 1652 : Lecture notes in artificial intelligence) ISBN 3-540-66325-8 CR Subject Classification (1998): I.2.11, H.2, H.3, H.4, H.5, I.2 ISBN 3-540-66325-8 Springer-Verlag Berlin Heidelberg New York This work is subject to copyright All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms or in any other way, and storage in data banks Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer-Verlag Violations are liable for prosecution under the German Copyright Law c Springer-Verlag Berlin Heidelberg 1999 Printed in Germany Typesetting: Camera-ready by author SPIN 10703969 06/3142 – Printed on acid-free paper Preface This volume contains the proceedings of the Third International Workshop on Cooperative Information Agents, held in Uppsala, Sweden, July 31 August 2, 1999 Modern information environments consist mainly of large, distributed and heterogenous resources which are connected via the Internet, Intranets, or virtual private networks These environments are open and may dynamically change over time To cope with such information environments means, in particular, to deal with uncertain, incomplete and vague information The e ective handling of uncertainty is critical in designing, understanding, and evaluating computational systems tasked with making intelligent decisions It is also crucial to the ultimate success and broad application of information agents on the Internet, as well as in any industrial context Moreover, any comfortable, reliable human-agent interaction via appropriate types of avatars, and the multi-dimensional representation of the information space available for individual users in the Internet, remains a challenging research topic The same holds for e cient service mediation by systems of collaborating information agents In particular, the research area of cooperative adaptive and mobile agents for information management on the Internet is still an uncharted territory in agent technology The interdisciplinary CIA workshop series covers the whole thematic range of intelligent and cooperative information agents Each workshop will focus on a few selected themes of particular relevance and current interest The CIA-99 workshop were built on the success of CIA-98 `Learning, Mobility and Electronic Commerce for Information Discovery in the Internet', LNAI Series, Vol 1435 , and CIA-97 `DAI meets Databases', LNAI Series, Vol 1202 Main topics of the CIA-99 workshop were mobile information discovery, advances in collaboration, mediation and negotiation, as well as personal assistance and human-agent interaction This workshop features 10 invited lectures and 16 regular papers selected from 46 submissions Acknowledgements First of all, we gratefully acknowledge the nancial support from our Co-Sponsors: Deutsche Telekom AG, Berkom GmbH, Berlin Germany , George Mason University, Fairfax, VA USA , Daimler-Chrysler AG, Research & Technology, Berlin Germany , Active Online Systems Ltd., London UK , and AgentLink ESPRIT Network of Excellence for Agent-Based Computing The workshop has been organized in cooperation with the special interest group on Distributed Arti cial Intelligence DAI of the German Computer Science Society and the Computer Science Department of Uppsala University Sweden Moreover, we thank all members of the program committee and all external referees for their very careful work in reviewing and selecting the contributions Uppsala, July 1999 Matthias Klusch, Onn Shehory, Gerhard Wei Program Committee Sonia Bergamaschi Wolfgang Benn Hans-Dieter Burkhard Brahim Chaib-draa Yves Demazeau Frank Dignum Innes Ferguson Klaus Fischer Christian Freksa Erol Gelenbe Carl Hewitt Mike Huhns Toru Ishida Leonid A Kalinichenko Bart Kosko Sarit Kraus H.-J Mueller Joerg P Mueller San Murugesan Pablo Noriega Moira C Norrie Aris Ouksel Mike P Papazoglou Amit Sheth Carles Sierra Elizabeth Sonenberg Kurt Sundermeyer Katia Sycara Peter Thomas Robert Tolksdorf Jan Treur Christian Tschudin Mike Wooldridge University of Modena, Italy Technical University of Chemnitz, Germany Humboldt University of Berlin, Germany Laval University, Canada Leibniz IMAG CNRS, France University of Eindhoven, The Netherlands Active Online Systems, London, UK DFKI German AI Research Center Ltd., Germany University of Hamburg, Germany Duke University, USA MIT AI Lab, USA University of South Carolina, USA University of Kyoto, Japan Russian Academia of Sciences, Russia University of Southern California, USA University of Maryland, USA Deutsche Telekom AG, R&D, Darmstadt, Germany John Wiley & Sons Corp., London, UK University of Western Sydney, Australia Institute for AI Research, Spain ETH Zurich, Switzerland University of Illinois at Chicago, USA Tilburg University, The Netherlands University of Georgia, USA CSIC AI Research Lab, Catalonia, Spain University of Melbourne, Australia Daimler-Chrysler AG, R&T, Berlin, Germany Carnegie Mellon University, USA UWE Bristol, UK Technical University of Berlin, Germany Vrije Universiteit Amsterdam, The Netherlands University of Uppsala, Sweden QMW College, London, UK General Chair Matthias Klusch DFKI German AI Research Center Ltd., Germany Co-Chairs Onn Shehory Carnegie Mellon University, USA Gerhard Wei Technical University of Munich, Germany Larry Kerschberg George Mason University, USA Organization External Reviewers Ismailcem Arpinar Paolo Ciaccia Ariel Felner Merav Hadad Catholijn M Jonker Ralf Kuhnel Francisco J Martin Anandeep Pannu Terry Payne Klaus Schild Leon Sterling Maksim Tsvetovat VII Table of Contents Information Discovery and Management on the Internet Invited Contribution: Agent Technology from a NASA Perspective W Truszkowski, H Hallock, and J Kurien USA Invited Contribution: Digital City Kyoto: Towards a Social Information Infrastructure 34 T Ishida, J.-i Akahani, K Hiramatsu, K Isbister, S Lisowski, H Nakanishi, M Okamoto, Y Miyazaki, and K Tsutsuguchi Japan Invited Contribution: Autonomous Search for Information in an Unknown Environment .47 E Gelenbe USA Invited Contribution: Resource Management in Agent-based Distributed Environments 61 A Brodsky, L Kerschberg, and S Varas USA, Chile Information Agents on the Internet Prototypes, Systems and Applications A Multi-Agent Architecture for an Intelligent Website in Insurance 86 C M Jonker, R A Lam, and J Treur The Netherlands Formation of Cooperative Behavior among Information Agents in Web Repository Change Monitoring Service 101 S Saeyor and M Ishizuka Japan GETESS Searching the Web Exploiting German Texts .113 S Staab, C Braun, I Bruder, A Dusterhoft, A Heuer, M Klettke, G Neumann, B Prager, J Pretzel, H.-P Schnurr, R Studer, H Uszkoreit, and B Wrenger Germany An Agent-Based System for Intelligent Collaborative Filtering 125 C O'Riordan and H Sorensen Ireland X Table of Contents Communication and Collaboration Inter-Agent Communication in Cooperative Information Agent-Based Systems .137 H Gomaa USA Intention Reconciliation in the Context of Teamwork: An Initial Empirical Investigation 149 D G Sullivan, A Glass, B J Grosz, and S Kraus USA, Israel A Similarity Evaluation Technique for Cooperative Problem Solving with a Group of Agents 163 S Puuronen and V Terziyan Finland, Ukraine A Computational Model for a Cooperating Agent System .175 S M Deen England Mobile Information Agents Mobile Agents Behaviours: From Declarative Speci cations to Implementation 196 C Hanachi, N Hameurlain, and C Sibertin-Blanc France Maintaining Specialized Search Engines through Mobile Filter Agents 208 W Theilmann and K Rothermel Germany Execution Monitoring in Adaptive Mobile Agents 220 W Vieira and L M Camarinha-Matos Portugal Mobile-Agent Mediated Place Oriented Communication .232 Y Kitamura, Y Mawarimichi, and S Tatsumi Japan Rational Information Agents for Electronic Business Invited Contribution Abstract : Agents and Electronic Commerce: Mechanisms and Protocols 243 M P Wellman USA 388 M Lewis tightly controlled and "intelligible" to humans If we can devise the means for the human to communicate her discriminations and semantic judgements to the agent it should be possible for the joint system to develop site specific information extractors fairly efficiently What is needed is some form of "programming by example" with provisions for indicating and discriminating among procedural instructions (accessing successive pages for instance), informational templates (the boundaries and constituents of classified ads or example) and string constants/variables for matching From the user's perspective, one would like to access a page through a browser, select (highlight) an instance of the items to be searched, and associate the selection's constituents with a schema for testing and extracting matching instances The design challenge is to build an interface which make this kind of direct manipulation specification as transparent as circling a classified ad and underlining an object's name and price Unlike the controlled experiments with TANDEM and MokSAF, our efforts in developing the InfoWrapper follow an iterative prototyping plan where through development and testing we explore a range of potential interaction schemes for bridging the gulf between what the user sees and what the agent parses The InfoWrapper has by far the most complex task of conveying human intent of the systems we have studied Like TANDEM, it must operate in an open environment where errors are likely and mechanisms for monitoring and evaluation must be designed in Like MokSAF, there are many possible ways to allocate tasks and control with good choices likely to emerge only after repeated testing We believe that the future of human agent systems will lie in such multi-method interactions which can combine demonstration, direct manipulation, machine learning, and command language in effective enough ways to bridge Norman's gulfs and allow communication of complex intents and perceptions Acknowledgements This research has been sponsored by ONR grant N-00014-96-1-1222 References P M Jones and C M Mitchell Human-computer cooperative problem solving: Theory, design, and evaluation of an intelligent associate system IEEE Transactions on Systems, Man, and Cybernetics SMC-25(7), pages 1039-1053, 1995 Anticipation Delegation, and Demonstration: Why Talking to Agents Is Hard 389 J Lee and N Moray Trust, Control Strategies, and Allocation of Function in HumanMachine Systems Ergonomics 35(10), pages 1243-1270, 1992 T Lenox, L Roberts, and M Lewis Human-Agent Interaction in a Target Identification Task In 1997 IEEE International Conference on Systems, Man, and Cybernetics, pages 27022706, Orlando, FL: IEEE, 1997 J.T Malin, D.L Schreckenghost, D.D Woods, S.S Potter, L Johannesen, M Holloway, and K D Forbus Making Intelligent Systems Team Players: Case Studies & Design Issues Human-Computer Interaction Design (NASA Technical Memorandum 104738) Houston, TX: NASA Johnson Space Center, 1991 B Muir Trust in Automation: Part I Theoretical Issues in the Study of Trust and Human Intervention in a Process Control Simulation Ergonomics, 39(3), pages 429-460, 1994 N Negroponte Being Digital New York, NY: Alfred Knopf,1996 D Norman Cognitive Engineering In User Centered System Design: New Perspectives on Human-Computer Interaction, eds D Norman and S Draper, pages 31-61, Hillsdale, NJ: Lawrence Erlbaum, 1986 E M Roth, J T Malin, J T and D L Schreckenghost Paradigms for Intelligent Interface Design In M Helander, T K Landauer, and P Prabhu (Eds) Handbook of Human-Computer Interaction, Second Edition, pages 1177 -1201, 1997 K Smith-Jentsch, J H Johnston, S C Payne (in press) Measuring team-related expertise in complex environments To appear in J A Cannon-Bowers and E Salas (eds.), Decision Making Under Stress: Implications for Individual and Team Training Wash., DC: American Psychological Association A Cooperative Comprehension Assistant for Intranet-Based Information Environments Ludger van Elst and Franz Schmalhofer German Research Center for Arti cial Intelligence (DFKI), University Bldg 57, Erwin-Schroedinger-Str., D-67663 Kaiserslautern felst, schmalhog@dfki.uni-kl.de Abstract A cooperative comprehension-assistant is described which represents text documents at three di erent levels of abstraction (surface, propositional and situational levels of representation) At the surface level texts are represented as bags of words without any linguistic structure For the propositional level a de nite but evolutionary growing set of predicates and concepts is used for selectively representing the most interesting parts of each text By a latent semantic analysis the situational level is calculated as the third level, which is a parsimonious representation of the complete contents of a document with respect to some previously established frame of reference (metric vector space) Together the three levels provide more exibility as well as precision for assisting human comprehension, document storage, retrieval and distribution Furthermore, it is shown how this comprehension assistant is applied for knowledge dissemination in a particular kind of intranet-based information environment (oligo-agent system) The individual and social components of knowledge sharing within such an environment are described by knowledge construction and knowledge integration processes To take the problem of long-term maintenance into account, the comprehension assistant and the oligo-agent system is embedded in a life-cycle model of cooperative knowledge evolution, consisting of a seeding phase, the system's evolutionary growth and periodical reseedings Introduction The successes of the internet as well as intranets have been tremendous There are now about 320 million pages in the World Wide Web On any given day a powerful search engine cannot even identify all these pages Only a relatively small portion (6 million) of these pages can be identi ed within a day With intranets there is the similar problem, that it is often quite di cult to nd the most relevant knowledge for a particular task at hand The large amount of electronically available information in combination with the continuously occurring changes of the stored information make the task of nding the best document available quite di cult Anyone who has frequently used the search engines on the internet has had the experience that the retrieved information did not satisfy his particular needs although the desired information was indeed available in the net: The retrieved information is often redundant and some portion of it may be quite irrelevant with respect to the particular query In order to alleviate this problem we have developed a cooperative comprehension assistant for a particular type of multiagent information system which was termed oligo-agent system 15] In this paper we describe the comprehension assistant which uses multiple levels of representations as has been suggested by theories of human cognition Thereafter it will be shown how this comprehension assistant can be utilized for the storage, distribution and retrieval of information in an intranet-based information environment Finally, a summary will be presented A Cooperative Comprehension Assistant 2.1 Knowledge Construction and Integration for Document Comprehension It is well known that one of the core tasks for systems of multiple agents (cf 1]) is to establish a shared context to allow for collaborative problem solving 9] This is especially true for human-centered approaches to knowledge management where the information landscapes consist of human and arti cial agents Here, a mutual understanding of the structure and contents of the organizational memory is central to enable the di erent actors to take over their speci c responsibilities and achieve an overall performance that is su cient for the respective needs of the multiple participants 11] We therefore employ a comprehension assistant within the social coordination region 15] This assistant was developed according to the principles of a cognitive theory of text comprehension, namely the CItheory 10], and yields knowledge representations of texts that are therefore relatively similar to the knowledge representations that humans form when they read a text Hence they are well suited for establishing mutual understanding To take the problem of long-term maintenance into consideration, this approach is embedded in a life-cycle model of cooperative knowledge evolution, consisting of a seeding phase, the systems evolutionary growth and periodical reseeding 8] As it is shown in Figure 1, two types of processes are employed to establish a shared context among the information providers and the information consumers by the comprehension assistant Whereas the various knowledge construction processes are quite general and therefore well suited for all the di erent participants (perhaps leading to redundant or inconsistent portions of information in some participants' views or with respect to speci c purposes), the knowledge integration processes yield a more coherent view and are therefore more closely tied to an individual user The construction processes are based on various speci c assumptions for the representation of the documents This frame of reference is negotiated between the participants in the seeding phase and should stay quite constant during the evolutionary growth where more and more users and documents participate in the system while other documents may be deleted and some users may no longer participate After a while, the chosen frame of reference, construction processes 3-level representation integration processes long-term query (demon) situational propositional sample of documents ranked list relevance computation of relevant documents surface ad hoc query (server) document document information provider evaluation & re-ranking consumer profile information consumer Fig Knowledge construction and knowledge integration processes for three-level representations as a means for accomplishing mutual understandings among information providers and information consumers i.e the "blueprint" of the organizational memory, may no longer be appropriate and the new representational assumptions must be negotiated (reseeding) Next we will describe the three-level representations of text documents with their respective construction processes in more detail 2.2 Three-Level Representations The most important task of the comprehension assistant is to represent text documents on a) a surface, b) a propositional and c) a situational level While the surface level is quite word-oriented, the two other levels utilize more abstract representation spaces (discrete spaces for the propositional level and continuous spaces for the situational level) surface level: On this level, documents are seen as "bags of words" Hereby we enforce an abstraction from the linguistic structure of the text as well as from concrete semantics of the words Technically speaking, the representation space is a highly dimensional vector space where the dimensions are all the words that occur in all documents Documents and queries are vectors in this space and the relevance of a document with respect to a speci c query is approximated by the similarity of the two vectors (e.g the cosine) Most search engines that are in use nowadays are based on this vector space model (e.g 13]) and the problems like low precision and recall values due to polysemy or synonymy are well known However, the method works very fast and can easily be understood by novice users as it doesn't require a deeper theory Furthermore, a good portion of practical problems can be solved with the assistance of this level, as the application of WWW search engines shows The only arrangements that have to be made in the seeding phase is the speci cation of parameters like stop word lists, the concrete weighting schema and similarity function Much past research has be done in this area 5] propositional level: This level of representation abstracts from the speci c words to diminish some of the problems of the surface level On this level, documents are represented in a discrete representation space that is based on a xed vocabulary which has to be negotiated in the seeding phase On the basis of a pre-de ned list of propositions which may consist of concept terms and/or predicate argument terms, each sentence will be searched for these speci c propositions Technically speaking, this amounts to searching for speci c terms or combinations of terms in a sentence Thereby synonyms will be matched to the same concept or proposition Such propositions are called CL-Propositions which stands for "controlled language propositions" These CL-Propositions have the following basic properties: { { { { CL-Propositions are predicates that contain no variables Predicates must be elements of a controlled language, i.e elements of an emerging ontology The terms of the controlled language as well as strings can be used as arguments CL-Propositions can be related to various segments of a document, e.g sentences or paragraphs Here, the choice of the controlled language is pivotal Basically, CL-propositions are normal forms of selected sentences By the application of some prototypes of CL-propositions and a thesaurus that maps the words of the documents to the controlled language selected sentences may be represented in this controlled language The number of prototypical CL-propositions may be at rst small, but with the use of the system it may incrementally grow to larger numbers (for a more detailed description see 7]) situational level: The situational level of a document will be generated as a vector in a latent semantic representation space 12] Alternative situation models of a text may thereby be constructed by using alternative semantic spaces For example, a text which describes a computer science department of an American university may be represented in the space of academic department descriptions of all academic disciplines in the whole world or in the more detailed representation space of computer science departments in North America The selection of a representation space thus determines the scope and the frame of reference 4] for the interpretation of a text old_doc term vector document folding-in abstract doc vector new_doc term_vector (sample) term-docmatrix (full space) Dn SVD reduction of dimensions abstract space folding-in Sim Cos Sn Rel Cos Tn folding-in query terms abstract query vector query vector example document Fig Technical speci cation of the construction processes at the situation level Figure summarizes the technical description of the situation level Of central importance for the situational level is the abstract space which is shown in the center of the gure This abstract space is obtained from the standard term-document matrix 13] by a singular value decomposition 6] which results in a reduced but still high dimensional semantic space that is represented by three matrices (D, S, T) These matrices contain the row and the column entities (matrices T and D) of the original term-document matrix as well as scaling values (diagonal matrix S) in such a way that by multiplying D, S and T the original matrix is approximated Here, an abstraction is achieved by the reduction of dimensions As is shown at the top region of Figure 2, old as well as new documents can subsequently by represented by abstract document vectors (folding in) As is shown in the bottom region of Figure 2, any user query (in the form of some sequence of search terms of the length i, i m, or in the form of some example document) may also be folded into the abstract representation space and thereby yield an abstract query vector For any given query, the most relevant documents (on the basis of the abstract semantic space) may now be determined by some similarity measure (e.g SimCos, the angle between the abstract document vectors and the abstract query vector) which is de ned in the abstract space The three levels thus allow a user to search for documents in a way that is adjusted to his prior domain knowledge We will now explain the usefulness of the three levels of representations by an example: Suppose, there is an intranet with a web-page for each computer science course that has been o ered by any instructor worldwide Furthermore, there are requests by a large number of international students for acquiring some speci c computer science training As the terminology is quite varied across di erent countries, a broker that uses only the surface forms of the course descriptions and the student request will not necessarily nd the best match For example, in some contexts, computer science, informatics, and electrical engineering may indeed be treated as synonyms (When a university does not have a computer science department, a true computer science course may be o ered by the department of electrical engineering) Fig Six hypothetical web-pages about academic classes As a minute example, imagine that there are n=6 courses o ered, where each course is described by a web page, as is shown in Figure Furthermore, there are high school graduates who are interested in studying arti cial intelligence and not yet have much knowledge about this eld When using a search engine with the search key "arti cial intelligence" they would nd only page When using the search key "computer science", they would nd page With such a surface oriented search, the users are literally getting what they are requesting With the three-level representation and the described comprehension assistant, on the other hand, these users may nd what they really wanted, even when they are not knowledgeable enough to explicitly request it In the information environment of the oligo-agent system for course brokering, the propositional level would encompass concepts like "engineering", "humanities" and "natural sciences" At the propositional level, "electrical engineering", "computer science" and "information science" would therefore all be mapped onto "engineering" In the latent semantic space of the situational level, all web-pages about science courses would be represented by vectors in the same vicinity of the metric vector space Web pages about humanity courses would similarly build a cluster, however, at a quite di erent location of the latent semantic space With the comprehension assistant, the high school graduates would again start out by using "arti cial intelligence" as search key With the retrieved page 4, there would also follow the propositional unit "engineering", which would subsequently retrieve the pages and 3, but not page (nor pages and 6), because mathematics is not mapped onto "engineering" at the propositional level However, when the search is extended to the situational level, all pages about science and engineering are found (1, 3, 4, 5), because with the latent semantic analysis these pages cluster in the same region of the multidimensional space, just like all humanity courses cluster in some other region of the space of university courses Courses and are therefore not retrieved at the situational level In summary, if a user has little or no knowledge of what he is really searching for, he may use some key terms and the surface level is then applied as in any other search engine As retrieval result he will, however, also obtain the propositional and situational representations of those documents as well In subsequent searches he can now use those propositional or situational levels of the documents that most closely correspond to his interests as a search key The search for relevant text documents over time thus advances from a super cial level (surface search) to a deeper level of understanding (latent semantic space) and thereby to a mutual understanding between user and system representations Figure also shows how the comprehension assistant can function as a tool for mutual understanding and cooperative comprehension among information providers and information consumers This is accomplished by the various knowledge construction and knowledge integration processes which are depicted in Figure The di erent roles of the participating agents can lead to quite ictive goals For example can the structure of an information archive that is \optimal" for the librarian be very di erent from the structure that is \optimal" for the distributors Therefore, the various frames of reference (or representational assumptions) that build the basis of the comprehension assistant must be negotiated between the agents An initial coordination of the individual and the global concerns of the di erent users is already achieved at the de nition time of all potential communications within the intranet This coordination is accomplished by a more or less representative sample of documents from which a three-level representation and the associated relevance scores are established Informationidenti cation is initiated by long-term queries as well as ad hoc queries The most relevant information (i.e the right documents) become explicitly speci ed (information identi cation time) Integration processes 7] form consumer-centered views (e.g individual re-ranking) and allow an information consumer to converse about his individual relevance rankings of the various selected documents on the basis of the three-level representation (cf 16]) Figure shows the particular in- terface which is used for conversing about the individual relevance of documents Through the various sliders and buttons, the parameters of the integration processes (e.g the weighting of di erent topics or the desired level of abstraction) can easily be manipulated and the e ects can be seen immediately Fig A dynamic query interface for conversing about the personal relevance of documents It should be noted that the comprehension assistant does in no way replace the human comprehender Quite to the contrary, the comprehension assistant and its human users are cooperating with one another and thereby enhance but also somehow change the human comprehension process This is comparable to the changes which are occurring in text composition when an author switches from a manual typewriter to a word-processor with spelling and grammar checkers As is well known, certain qualities of the task execution are modi ed by using the new technology 2] Embedding in a Multi-Agent System Figure shows how the described comprehension assistant is embedded in an oligo-agent system An oligo-agent system is a special kind of multi-agent system, where 1) there is only a small number of di erent types of agents and 2) the agents are de ned by their social responsibilities for achieving and maintaining certain qualities of the whole agent-system and the particular information environment in which they operate (cf 15]) Inter-/Intranet ORGANIZATIONAL REGION publishes documents provides documents and basic services GloMa distributes documents forwards documents for distribution forwards classified documents SOCIAL REGION forwards documents Comprehensionbased Assistant ProPersonA provide and describe documents Producer/ Distributor requests documents ConPersonA present and converse about documents INIDIVIDUAL REGION Information Consumer Fig The structure of the oligo-agent system for distribution and comprehension assistance within the organizational memory We will now explain how the comprehension assistant is utilized for the various distribution- and comprehension-based tasks which have to be performed within an intranet At rst we have to explain some additional speci cations Repositories of potential information consumer groups and document groups as well as templates of distribution and interests-lists are held in a relational database system (RDBS) Based on these repositories, speci c distribution and interests-lists are constructed (or chosen) and re ned These tasks also specify information as to when the information should be delivered A demon acts upon this information Whenever a speci c distribution or interests list reaches its distribution time this demon computes the delivery information which basically consists of a table of (user, document info)-pairs This table is processed by a standard delivery mechanism (e.g e-mail), sending the document noti cations to the user's inbox The delivered information is ltered by the consumer agent (ConPersonA) and thus categorized individually Each category has presentation time information which is used by a ConPersonA-demon to initiate presentation via the special user interface that is shown in Figure and allows for individualized ranking and selection of documents How the document providers and information consumers of the intranet can employ the distribution and comprehension assistant to improve their consensual understanding of documents in the organizational memory is shown in the form of a use-case diagram (see Figure 5) A document provider (who may be either an author or a distributor) can publish a document in the organizational memory in one of two ways He may have his own ideas as to where in the intranet (i.e the virtual library) the document should be stored and to whom it should be distributed In this case he would directly forward the documents and/or distribution list to GloMa, the global manager that has access to and maintains a representation of the organizational memory Alternatively, he can call on the comprehension assistant, that will then make suggestions as to where the document is to be stored and who should be informed about its publication In either case, GloMa will eventually store the newly published information in the organizational memory and distribute the information via the personal agents to the potential information consumers In addition to this push-oriented approach to knowledge management, the integrated assistant also provides a sophisticated information pull solution that is based on content and meta-content descriptions of the documents as well as user interests With the personal agent and the described dynamic queries interface (see Figure 4), a user can specify his interests to the comprehension assistant, that will in turn communicate to GloMa which will then provide the speci c documents Summary The described cooperative comprehension-assistant that is to be applied for knowledge dissemination in an organization (i e the right documents to the right people at the right time) can be more generally characterized through the following design speci cations and accomplishments: Based on theories and models of human text comprehension 10], 14], a fully automated comprehension assistant with three levels of representations is used to achieve mutually shared representations between the oligo-agent system and their users With increasingly more people applying this system, the mutual understanding may be shared among increasingly more people and increasingly more documents Individual and global concerns are coordinated in an oligo-agent system with shared responsibilities While the individual agents are client-based, the global agent runs on an application server GloMa uses an RDBS to keep and maintain repositories that are used for the de nition of user pro les (information consumers), document groups and distribution and interests-lists Thereby it is possible to use the available information, consisting of document attributes, document contents and organizational structures as a whole Concerning the distribution and interest identi cation tasks, GloMa maintains a continually updated representation of the information environment at the relevant level of abstraction Decoupling of de nition time, information-identi cation time and presentation time allows distribution to be determined either by the individual or as a common responsibility of administrator, author, distributors and information consumers These responsibilities may concern one-shot distributions as well as periodical repetitions of some general distribution speci cation From an application point of view, the functions that are provided by the comprehension assistant and its embedding in the oligo-agent system can be compared to the Fishwrap system 3] The information consumer gets a personalized view of the information environment This view consists of individual aspects that are based upon a semantic document analysis as well as upon organizational aspects (distribution lists), in which a distributor determines the portion of information that is delivered to the consumer However, unlike an individualized newspaper, the proposed system has a very sophisticated representation about the ripeness and expiration time for relevant information This allows for a di erent "date of publication" for each "line" of the newspaper The combination of these elements leads to a exible tool for handling di erent aspects of information distribution and information utilization which are the core problems of document and knowledge exploitation in the knowledge society References Ambite, J L., and Knoblock, C A.: Agents for Information Gathering, IEEE Expert, September/October, 1997 Boy, G A.: Cognitive Function Analysis NATO-ONR Workshop Presentation, Washington D.C., 1997 Chesnais, P R., Mucklo, M J, and Sheena, J A.: The Fishwrap Personalized News System http:// shwrap-docs.www.media.mit.edu/docs/dev/CNGlue/cnglue.html, 1997 Clancey, W J.: Situated Cognition: On human knowledge and computer representations New York: Cambridge University Press, 1997 DARPA (Ed.): Proceedings of the 6th Message Understanding Conference (MUC-6), Columbia, Maryland, November 6-8, 1995 Morgan Kaufmann, San Francisco, 1996 Deerwester, S., Dumais, S., Furnas, G., Landauer, T., and Harshmann, R.: Indexing by Latent Semantic Analysis Journal of the American Society for Information Science, 41(6), 391{407, 1990 van Elst, L.: Ein kooperativer Informationsassistent zum gemeinsamen Verstehen von Textdokumenten (An information assistant for the cooperative comprehension of text documents ) Master Thesis, Department of Computer Science, University of Kaiserslautern, 1998 Fischer, G., McCall, R., Ostwald, J., Reeves, B., and Shipman, F.: Seeding, Evolutionary Growth and Reseeding: Supporting the Incremental Development of Design Environments In Proceedings of ACM CHI94 Conference on Human Factors in Computing Systems, 2, p 220, 1994 Fischer, G., and Redmiles, D.: Enhancing Indirect Long-Term Collaboration with Intelligent Agents NSF-Grant IRI-9311839, Boulder 10 Kintsch, W.: Comprehension: A Paradigm for Cognition Cambridge University Press, 1998 11 Krauss, R M., and Fussell, S R.: Mutual knowledge and communicative e ectiveness In J Galegher, R Kraut, C Egido (Eds): Intellectual Teamwork: Social and technological foundations of cooperative work, pp 111{146 Hillsdale, New Jersey: Lawrence Erlbaum Associates, 1990 12 Landauer, T.K., Foltz, P.W., and Laham, D.: Introduction to Latent Semantic Analysis Discourse Processes, 1998 13 Salton, G.: The Smart Retrieval System Experiments in Automatic Document Processing Prentice Hall, 1971 14 Schmalhofer, F.: Constructive Knowledge Acquisition: A Computational Model and Experimental Evaluation Mahwah, N.J.: Lawrence Erlbaum Associates, 1998 15 Schmalhofer, F., and van Elst, L.: An oligo-agent system with shared responsibilities for knowledge management In D Fensel, and R Studer: Proceedings of EKAW '99, Springer-Verlag, 1999 16 Williamson,C., and Shneiderman, B.: The Dynamic HomeFinder: Evaluation dynamic queries in a real-estate information exploration system Proceedings of the ACM SIGIR'92 Conference, Copenhagen, Denmark, (June 1992), pp 338{346, 1992 Author Index Akahani, J.-i 34 Banerjee, S 265 Benn, W 335 Braun, C 113 Brodsky, A 61 Bruder, I 113 Camarinha-Matos, L M 220 Deen, S M 175 Dusterhoft, A 113 Elio, R 347 Elst, L van 390 Fankhauser, P 323 Gelenbe, E 47 Glass, A 149 Gomaa, H 137 Gorlitz, O 335 Grosz, B J 149 Haddadi, A 347 Hallock, H Hameurlain, N 196 Hanachi, C 196 Heuer, A 113 Hiramatsu, K 34 Isbister, K 34 Ishida, T 34 Ishizuka, M 101 Jonker, C M 86 Kashyap, V 303 Kerschberg, L 61, 265 Kitamura, Y 232 Klettke, M 113 Kraus, S 149 Kurien, J Lam, R A 86 Lam, K Y 291 Langley, P 347 Lewis, M 365 Lima, T 303 Lisowski, S 34 Mawarimichi, Y 232 Miyazaki, Y 34 Nakanishi, H 34 Neubert, R 335 Neumann, G 113 Okamoto, E 291 Okamoto, M 34 O'Riordan, C 125 Papazoglou, M P 245 Prager, B 113 Pretzel, J 113 Puuronen, S 163 Rothermel, K 208 Saeyor, S 101 Schmalhofer, F 390 Schnurr, H.-P 113 Sheth, A 303 Sibertin-Blanc, C 196 Sorensen, H 125 Staab, S 113 Studer, R 113 Sullivan, D G 149 Tatsumi, S 232 Terziyan, V 163 Tesch, T 323 Theilmann, W 208 Thompson, C 347 Treur, J 86 Truszkowski, W Tsutsuguchi, K 34 Uszkoreit, H 113 Varas, S 61 Vieira, W 220 Wang, X F 291 Wellman, M P 243 Wrenger, B 113 Yi, X 291 Zhang, C Q 291 ... environment Mental Models of Subsystems and Instruments Mental Models of Model Integration, Operations, Collaborations, and Model Evolution "Lights Out" System A 1, A 2, , Ai ES 1, ES 2, , ESi Other... subsystem has responsibility for executing the maneuvers via commands to the spacecraft's thrusters, which also at times may be used for attitude control and momentum management 2. 2 .2 Executive, Command... Vieira and L M Camarinha-Matos Portugal Mobile-Agent Mediated Place Oriented Communication .23 2 Y Kitamura, Y Mawarimichi, and S Tatsumi Japan Rational Information Agents for
- Xem thêm -

Xem thêm: Cooperative information agents III, , matthias klusch, onn m shehory, gerhard weiß, 1999 2 , Cooperative information agents III, , matthias klusch, onn m shehory, gerhard weiß, 1999 2

Mục lục

Xem thêm

Gợi ý tài liệu liên quan cho bạn