We are now mid-way through the UMDL project. At the start of the project, one of our objectives was explore various ways of defining, building, evaluating, and deploying digital libraries. Towards this objective, we created an agent architecture that served as the basis for a deployable testbed. This architecture was populated with a variety of agents that provide services, content, and user interface.
In the course of building and researching digital libraries over the past two years, we have gained both experience and insight into what digital libraries are all about. We have spent some time over the past several months evaluating what we have accomplished, where we have fallen short, and most importantly, where we need to go in the remaining two years. The plans for the coming year reflect some of this evaluation.
Architecture and Agents. Teaming between specialist agents to provide goods and services will only work if agents can extract from their interactions suitable benefits. In turn, this requires them to be capable of modeling each other sufficiently to predict what it will take to convince others to participate, and how to avoid participating with agents that have previously been detrimental. We have been developing general mechanisms that can be integrated into the decision processes of agents that allow them to construct models of others and to use the models for strategic reasoning, so as to participate effectively in the open UMDL system where benevolence among agents cannot be assumed.
Some of these mechanisms are being integrated in our ongoing efforts to incorporate auction agents into the current UMDL agent teaming process for query processing. Whereas previously there were few alternative agents to engage in a team, we have been introducing competition among agents of a particular class (task planning agents), and exploring the use of auction agents for effective task allocation. Part of this process involves the task planning agent sending offers to the auctions expressing how much their services currently cost. Determining this cost is one place where our strategic reasoning mechanisms are being manifested.
Intellectual Property and Economics. At the May DLI meeting, we demonstrated the first market-driven negotiation scenario in UMDL. To support this scenario, we designed and implemented (1) market facilitator agents (UMDL auctions), and (2) bidding strategies for some new and existing agent types. Since May, we have augmented our infrastructure for market-based negotiation by developing a general auction manager, and specifying goods and bidding behavior for broader classes of UMDL agents and services.
Advanced User Interface. Our research so far has focused on developing a deep understanding of information gathering and sensemaking activities. As such, we have been examining a broad variety and broad temporal scope of tasks. The users and their tasks range from post-docs developing research programs, to students buying a CD player, to people looking for jobs, to a senior editor creating a weekly newspaper.
The goal of such a broad and deep understanding of users and tasks is the presumption that higher level human goals will appear generally across these activities, and noting those will help us understand the structure of each individually. Furthermore it has moved us from just examining the querying, or even just the finding behavior of our subjects, rather whole embedding super-task
Each team member has completed a number of interviews and has submitted notes to the group which we discuss in depth, seeking parallels and divergences between cases, with results informing subsequent interviews. Our group analyses has focused in three primary areas: one, the creation of a dual conceptual structure for the overall information gathering and organizing tasks; two, the development of a corresponding dual conceptual structure for the procedures executed in undertaking the various tasks; three, a general understand the links between the task characteristics and the procedures followed to accomplish the tasks.
In parallel to our research in users and tasks, technology research has been focused on micro prototypes -- explorations of interface fragments that might be of use in an advanced information worker's environment. The micro-prototypes have been implemented using the PAD++ platform. The micro-prototypes which have been built include: various search+browse synergy demonstrations, graphical search histories, tools for early-structuring, shared workspaces, and webbook visualizers.
Architecture and Agents. We will work to complete the integration of the auction agent capabilities into the UMDL. From there, we can begin to experiment within the UMDL's information economy with the advantages and costs of strategic reasoning, along with techniques for the acquisition of agent models (and decisions about how deep those models should be).
Intellectual Property and Economics. Over the next six months, the IPE group will continue to focus on extending the market-based negotiation infrastructure. This includes design of a more flexible description language for library goods and services, implementation of a simple bookkeeping mechanism for agent accounts, and integrating our negotiation process with external electronic payment and exchange protocols.
Service Classifier Agent (SCA). A new Service Classifier Agent (SCA) is being developed to map complex descriptions of library services to labels used for indexing in a registry. Service descriptions and requests for services are coded in language defined by the UMDL ontology, and may include connectives and quantifiers yielding expressiveness equivalent to first order logic. The SCA will utilize Loom, a KL-ONE description logic system implemented in Lisp, to classify agent-types. When an agent advertises, a description of its capabilities is merged with existing agent-types in the SCA's taxonomy and associated with a label. When an agent requests a service, the SCA identifies all agent-types that can match the request, then the registry selects an individual agent. The SCA will contribute flexibility, scalability and extensibility to the UMDL architecture, by permitting the formation of agent teams to be based on rich but static descriptions of services which are independent of the particular constellation of agents available at some time. Thus, when a newly developed agent is advertised it will automatically be utiltized in appropriate ways without modification of previously existing agents. Currently the high level design of the SCA is complete, and formal specification of a description and query language (DQL) and implementation of the agent is underway.
At a general agent modeling level, we have been developing logical models of "trust" among agents. In other words, how can we model, from an agent (not system) perspective environments where not all agents are trustworthy. As an example: how should an agent act when the "trustworthiness" of an input request is in doubt. This new research effort builds upon work in security, negotiation, and agent modeling. Yet, it does not prescribe any particular mechanism for either determining the trustworthiness of an agent (a security concern), nor strategically interacting with an agent (a negotiation concern). Instead, the model is a framework for analyzing how an agent could use various security or negotiation mechanisms to reach its own ends.
Complementing the work on trust, we have been working on developing a model of agent 'commitment' inspired by Shoham's Agent Oriented Programming paradigm. This model is intended to show how a software agent makes decisions about which action it should commit to do at some future time. This commitment to action is in response to some event or message and must correctly reflect the agents goals and limited capabilities. In this model, no assumptions are made about what the mechanisms for generating possible commitments or how preferences over actions/outcomes are actually decided. Possible commitments can arise from negotiations, planners, economic models, etc. An agent selects from a set of possible commitments without generating conflicting commitments. The notion of retracting a conflicting commitment (where possible) gives added flexibility.These agent models of trust and commitment benefit developers of agents by giving them a concrete definition of the 'commitment' relationship that will occur among UMDL agents.
Ontology. The UMDL ontology has undergone considerable change the past six months. The scope of the ontology has broaden to include intellectual work and services. In addition to broading, the organization of the ontology has improved, and the number of concepts defined has increased.
We have translated portions of the ontology into an inferencing system. This system demonstrates how the ontology can be used to classify and search for various concepts. For example, we can use this demonstration system to search in more general ways than a relational database (e.g., we can look for "subsuming" relationships). This work has fostered the development of the Service Classifier Agent (described later in this document).
Advanced User Interface. Our future efforts will move to formalize our understand of users and tasks with a push towards design and implementation. In particular our next stage will be to draw design implications from our task analyses.
Based on the implications we draw, we will increase our micro-prototyping activities in a more targeted way, begin a program of user evaluation and testing, and begin modest linkages to the UMDL infrastructure, with beginning user interface agent (UIA) definitions.
The UMDL project has had some major successes. Figure 1 shows the agents in UMDL last year, while Figure 2 shows the agents in the system this year. There has been growth both in the number of agents and the complexity of interaction among them. This has the result of providing better services to the user. A premise of our agent architecture is that it is inherently more extensible than traditional client-server systems. This year, we were able to add a variety of new agents, providing new services, without negatively impacting the existing system. This is an indication of the extensibility we were hoping to demonstrate. On the education front ... ON the IPE front .. On the collection front ...
Figure 1. UMDL Agents 1995
Figure 2. UMDL Agents 1996
Please see the following reosurces:
Improvement of Design Documentation. (Goal 3.a.8). We continue to revise and update the UMDL architecture design documentation. Our primary efforts this term have been to provide detailed documentation of the Collection Search and Retrieval Protocols and the objects which participate in the protocol. The latest version of this document may be viewed at http://telemachus.engin.umich.edu/documents/umdl-impl.
Continued Enhancements To The Collection Search And Retrieval Protocol (Goal 3.a.9). The Collection Search and Retrieval (CollSR) Protocol is the same protocol that we use to achieve interoperability with other DLI sites. The CollSR/Interoperability protocol is built on top of ILU, Xerox's CORBA implementation. Due to delays in releasing adequate C++ support in the new version of ILU (which Stanford has already adopted), we have been unable to pursue further interoperability experiments with the other sites. However, we have continued to implement enhancements to the protocol for use within UMDL. For example, we have improved error handling to increase robustness and reliability. We are currently adding facilities to support persistence of search results for periods of time longer than a single search session. These improvements will translate into better interoperability once we are able to upgrade to the new ILU release.
Webbook Agent and Indexer (Goal 3.a.10). See IIc. Usage and User Base-- Information Gathering Environment (Goal 3.A.14), and New Java Interface to UMDL (Artemis) below.
Thesaurus Development (Goal 3.a.11). The BSO Agent has been generalized to a thesaurus agent, of which the BSO agent is an instantiation. The agent provides all previously supported functions. We have added term browsing capabilities. A prototype which uses this capability may be viewed at http://telemachus.engin.umich.edu/cgi-bin/UMDL/BSO. This prototype uses the same protocols which will soon be used by the User Interface Agent to provide search term browsing in Artemis.
Collection Query Transformation (Goal 3.a.12). With the development of our CIA for FTL collections, we have our first example of collection query transformation. We have successfully converted the same user query into formats for Z39.50 and FTL. The user query is sent to the CIA via the collection search and retrieval protocol. The CIA converts the query into a form that is understood by the collection it represents and passes the query to the collection's search engine.
New Java Interface To UMDL (Artemis). This Java-based interface has been designed based on analysis of user needs to be more than a simple query system; more explicitly than other searching tools, the new interface is designed to allow users to organize their thoughts and queries in a manner centered on their research interests (see Appendix A). We continue to work on the interface, refining some components to be more interactive, filling out the feature list, and updating to Java 1.0.2.
We have upgraded the User Interface Agent to have the capacity to maintain workspace information for the user. This includes features to save old search results and user-constructed folder hierarchies to organize individual results. To access these features via the web, we have built a Java-based API. More recently we have been working on a redesign and reimplementation of these systems to improve stability and responsiveness.
CIA for FTL Collections. We have developed a CIA to provide access to our FTL collections. The CIA uses the Collection Search and Retrieval Protocol to submit queries and provide access to result sets. Our FTL collections include Elsevier Earth and Space Sciences Journals, Tulip, and UMI periodicals. They are full-text searchable and provide full bit-mapped page images of articles.
A project meeting was held on July 7 in New York with both current and potential publishing partners. Representatives from McGraw-Hill, Cambridge University Press, Springer-Verlag, the Institute for Scientific Information (ISI), Grolier Interactive, and Elsevier Science attended the session. We reviewed the current status of the UMDL architecture, collection development, and commerce infrastructure. The open discussion focused on issues of special concern to the publisher partners, such as intellectual property management and pricing, and support for developing collection interface agents that accurately represent the content of publishers' collections.
Negotiations continue with Academic Press and ISI to provide additional content for UMDL. We are hoping to acquire a significant body of journal literature from Academic Press, and major citation indexes from ISI.
Web Document Search and Retrieval (Goal 3.a.10). We developed a Web-Book Agent in the UMDL Agent Architecture. Web-books found in this way were then registered in the UMDL Conspectus and were searchable by our a local search engine. Our intent in creating this agent was twofold: explore how third-parties can develop agents, and create a unit larger than a webpage to register in the Conspectus. With the respect to the former goal, we found that indeed, those not directly related to the UMDL project can successfully develop agents and integrate them into the agent network. This is an important finding: if the UMDL is to be richly populated with agents that can provide a broad-range of services, then third-parties must be involved in that effort. And, while our heuristic for identifying a web-book (i.e., look at the core name) is simplistic, it can successfully be used to create website collections that can be registered and subsequently searched.
Information Gathering Environment (Goal 3.A.14). We have designed and are implementing an interface to the UMDL that should truly support the query and re-search activity involved in inquiry-based science projects for high school (and middle school) students. "Artemis." (Goddess of the Hunt) has explicit "scaffolding" to support students as they learn the science content (earth and space sciences) as well as skills to carry out an inquiry (see Appendix A).
Add New York Public Library And Stuyvesant High School As Testbed Sites (Perform Baseline Evaluation At These Sites) (Goal 3.B.3). Inasmuch as the UMDL is not ready to be deployed to non-local sites, it will continue to be hard to engage organizations to participate in this project. Thus, while we have been in discussions with the New York Public Library (e.g., the head of the newly opened Science, Industry, and Business Library), we are in a holding pattern until we have a more stable system to offer them. That said, we are also in discussions with a number of non-local sites who have expressed great desire to have access to the UMDL. Thus, once we are further along with deployment, we will be better able to work with sites such as NYPL.
Deploy Release 2 at Test Sites (Goal 3.B.4). We decided to not deploy the UMDL in the participating high school classrooms this past Winter semester. There are enough weaknesses with the current system (e.g., needs more student-appropriate collections, needs a more supportive interface) that we felt that high school students would not find it sufficiently compelling and effective. Rather, we endured the use the Internet as a vehicle for exploring how to use online resources. (We found, for example, that students need a great deal of assistance in formulating a driving, researchable question and in formulating and reformulating keyword searches. The new interface, Artemis, expressly deals with this need.) We further explored the motivational and academic impact of having students actually publish their products and reports on the Internet. (We found, for example, that this publishing activity did result in a greater percentage of students than normal engaging in serious scientific and intellectual effort. Again, Artemis has been explicitly designed to make the publishing and sharing of student findings' straightforward.)
Development continues with the PAD++ on campus prototype. Faculty have been interviewed concerning task analysis and current tool use, and will be involved in prototype evaluation in the Fall and Winter semesters. Initial evaluative deployment of a prototype on a very small scale will take place in Winter 1997. Librarians are currently supporting resource identification and being kept informed of the development of the prototype and other UMDL initiatives.
Evaluate Precision Of Search Results At Testbed Sites Against Different Search Engines And Against Linguistic Algorithms (Goal 3.C.1). This goal has been subsumed by the user-centered approach to interface design, where timely and relevant results are critical in the fast-paced school environment.
Evaluate UMDL Release 2 at Test Sites (Goal 3.C.2). Over 1,000 middle and high school students in Ann Arbor used our online curriculum units this past semester. Inasmuch as this was our first significant use of online materials, and since we were using the Internet and not the UMDL itself, our evaluation focused on the strengths and weaknesses of (1) our curriculum materials, (2) our teacher development strategies, (3) instructional strategies employed by the teachers, (4) student search and inquiry strategies, (5) and the student publishing strategies. We highlight this formative evaluation in a paper just published in the Proceedings of the International Conference of the Learning Sciences (Northwestern University, July 1996).
Evaluation at UM campus hinges on the evaluative deployment and testing of the PADD++ prototype, the development of which is being led by Professor George Furnas. Testing of the interface, design, performance, and theory of development of the prototype will be examined.
Develop Curriculum Units, Register Appropriate Collections, And Develop Instructional Strategies For UMDL Use. Based on our experiences this past year in the high schools in Ann Arbor, we are in the process of revising are online curriculum materials (e.g., students tend to not read the materials, and thus we are moving to trying convey information in a more visual, direct manner). Moreover, we are working closely with the participating high school teachers to develop effective strategies for dealing with the wide range of issues involved in having students carry out science inquiry using online resources (e.g., how to help students stay engage over a number of days on a multi-day investigation, how to help students reformulate their driving questions and online searches as a function of they find over the multi-day investigation). And, we have mounted a major collection building effort: while some of the material provided by our publishing partners are relevant to high school science, we need to augment that material with WWW material (and better register all materials so as to make them more findable).
Deploy UMDL with Artemis at Ann Arbor High Schools. By winter of 1996, Artemis will be deployed as the front end to the UMDL. Moreover, due to a significant collection building effort during the summer and fall of 1996, there will be a rich set of collections represented in the UMDL Conspectus. Students, therefore, should be able to rely solely on the UMDL for their online resources.
Evaluate Impact of UMDL/Artemis. Inasmuch as the UMDL/Artemis explicitly scaffolds the inquiry process and provides a rich set of collections from which to draw, we expect that the high school students should be able to be more effective in their inquiry activities. We should see this impact in both process measures (e.g., amount of time spent online searching vs. time spent constructing an argument) and product measures (e.g., quality of final report). We have some baselines from this years evaluation effort that we may be able to use for comparative purposes.
Populate the UMDL with Supportive Agents. We are finding that we need a range of services to support collection building, collection search and rating, etc. We will continue to work at populating the UMDL with agents that support these services. This activity both makes our activity more effective, as well as exploring the mechanisms that need to be in place to support such agent construction.
There have been a few personnel changes in the past six months. JoAnne Kerr has joined the UMDL team as an Administrative Assistant to Dan Atkins, and coordinator of the UMDL project. JoAnne joined the UMDL team in August.
While there have been no changes or additions in faculty, the mix of students involved in the project have changed. Peter Weinstein and Yuhua Liu have joined the project. Both are working on the UMDL interface design. Finally, there are no longer undergraduate students working on the project.
1/9 Dr. Helen Shen, Hong Kong University of Science and Technology
Dan Atkins, Bill Birmingham, Randy Frank, Wendy Lougee, Greg Peters
1/9 Bob Rossbach, WTVS Detroit (television station)
Wendy Lougee, Elliot Soloway
1/10-11 Defense Information Enterprise Technology (DIET) Program PI meeting
Sunnyvale, CA
Bill Birmingham
1/16-19 Association for Library and Information Science Education Annual Conference
San Antonio, Texas
Dan Atkins, Olivia Frost, Laurie Crum
1/19-25 ALA Midwinter
San Antonio, Texas
Dan Atkins, Olivia Frost, Laurie Crum
1/29-2/2 Patrick Finn, La Plaza de Taos
Semnar on the Le Plaza de Taos (community network)
Dan Atkins, Joan Durrance
2/ MAME (Michigan Association of Media in Education)
Grand Rapids, Michigan
Keynote Presentation: Elliot Soloway
The Role of Digital Libraries in Supporting Inquiry Learning
2/1 Carlene Ellis, Intel
Dan Atkins
2/5 Roland Dietz, Senior VP, Elsevier Science
Wendy Lougee, John Price-Wilkin, Dave Rodgers
2/9 Ontario Library Association
Toronto, Canada
Speaker: Dan Atkins
2/16 Carol Moore, Director University of Toronto Library
Wendy Lougee, John Price-Wilkin
2/26 Sang-Wan Han, Sung-Hyuk Kim, Sung-Bin Moon
Korea
Dan Atkins
2/29 Academic Press Library Advisory Group
Orlando, FL
Wendy Lougee
3/3-8 OMG (Object Management Group)
David Richardson
3/4-5 ERCIM meeting
Nice, France
Bill Birmingham, Kathy Willis
Presentation: Bill Birmingham
3/5 Committee on Institutional Cooperation, Chief Information
Officers and Library Directors
Wendy Lougee, Randy Frank, John Price-Wilkin
3/ Rank Xerox
Grenoble, France
Bill Birmingham, Kathy Willis
3/5-7 Netscape Developers Conference
Greg Peters
3/8 George Shipman, University of Oregon
Dan Atkins
3/15 School of Information is officially launched
3/15 MACUL (Michigan Association of Computer Uses in Learning)
Lansing, Michigan
Keynote Presentation: Elliot Soloway
The Role of Digital Libraries in Supporting Inquiry Learning
Speaker: Raven Wallace
3/18 National Digital Library Federation, Washington, DC
Wendy Lougee
3/19 Research "Science Fair"
Capitol Hill, Washington
Raven Wallace
3/19 Ken Perlin, Director, Center for Advanced Technology in Digital Multmedia
Media Research Laboratory, New York University
AI Seminar: Two research Projects: PAD and Improvisational Animation
3/20 Jerry Butters, A T & T
Wendy Lougee, Randy Frank
3/21-22 7th Annual IPoSCE Review
The University of Michigan
Ann Arbor, MI
Speakers: Jose Vidal (Building and using agent models in an agent-based
digital library); Tracy Mullen (Exchangng goods and services in a digital
library); Ansoara Nica (Resource discovery in large-scale distributed library
systems)
3/26-4/5 Joerg Haake, Anja Haake
GMD, Darmstat
UMDL team
Seminar (4/4/96): Hypermedia-based Collaboration Support
3/31-4/1 OCLC Authority Control Conference
Dublin, Ohio,
Speaker: Karen Drabenstott
"Beyond Online Catalogs: The Role of Authority Control in Digital Libraries
4/1 Leif Hansen, Copenhagen Business School
Wendy Lougee, John Price-Wilkin
4/1 Chuichi Tanaka, Takanori Hayashi, Hatsuhito Mitsuhashi
Japanese Ministry of Agriculture, Forestry and Fisheries
Dan Kiskis, Wendy Lougee, John Price-Wilkin
4/2 IBM Steering Committee
Speaker: Dan Atkins
4/14-15 Amy Friedlander, Editor, D-Lib Magazine
UMDL Team
4/14-18 CHI '96: Common Ground
Vancouver, British Columbia, Canada
George Furnas
4/14-18 NAB '96
Las Vegas, NV
Bill Birmingham
4/22 Sue Corbett, Blackwell Publishers
Wendy Lougee, John Price-Wilkin
4/25 W3C Workshop on High Quality Printing from the Web
Spencer Thomas
5/ Advanced Digital Library Conference
Invited Panel on The Role of Digital Libraries in Education
Library of Congress
Elliot Soloway
5/2 Lu Kabir, Oracle
Wendy Lougee, Randy Frank
5/5 Tom Kalil, White House
Dan Atkins, Mike Wellman, Wendy Lougee, Randy Frank
5/6 NSF/ARPA/NASA Site Vist
Ann Arbor, MI
UMDL Team
5/6-7 Les Gasser, NSF
UMDL Team
5/8 Maria Luisa Arenas-Franco, Pantificia Universidad Catolica de Chile
Wendy Lougee
5/16-17 All Project Digital Library Initiative Workshop
University of Michigan
Ann Arbor, MI
UMDL Team
5/22 Media Union Dedication
5/31 Mark Abel, Intel
Dan Atkins
6/3 Liz Turgeon, Addison Wesley
Dan Atkins
6/3 James Wells, Real Video
Dan Atkins
6/10-14 CARIS '96
Federicton, Newe Brunswick
Bill Birmingham
6/13 Lorrin Garson and additional representatives
American Chemical Society
Dan Atkins, Wendy Lougee
6/-22 E D - M E D I A & E D - T E L E C O M 96: World Conference on
Educational Multimedia and Hypermedia and World Conference on
Educational Telecommunications
Boston, MA
Elliot Soloway, Invited Speaker
Title: The Role of Digital Libraries in Learning and Teaching
6/25-28 INET '96
Montreal, CANADA
Jeff MacKie-Mason
6/28 Alexandra Jankovich, Elsevier Science
Jeff MacKie-Mason
7/4-10 American Library Assocation Annual Meeting
New York
Dan Atkins and other members of the UMDL team
7/7 Publishers Meeting
Waldorf Astoria Hotel, New York
Dan Atkins, Wendy Lougee, Mike Wellman
7/11 Dr. Sajjad Ur Rehman, Department of Library and Information Science
International Islamic University, Kuala Lumpur
UMDL Project Staff
7/11 Dr. Ahmet Celik, Hacettepe University, Turkey
UMDL Project Staff
7/26 Kakugyo S. Chiku and Mr. Moroya
Kanazawa Institute of Technology Library Center
UMDL Project Staff
V. Papers
Alloway, A., Bos, N., Hamel, K., Hammerman, T., Klann, E., Krajcik, J., Lyons, D., Madden, T., Margerum-Leys, J., Reed, J., Scala, N., Soloway, E., Vekiri, I., Wallace (1996) Creating an Inquiry-Learning Environment Using the World Wide Web, Proceedings of the International Conference of the Learning Sciences, Northwestern University, Evanston, IL
Atkins, D.E., Birmingham, W.P., Durfee, E.H., Glover, E., Mullen, T., Rundensteiner, E.A., Soloway, E., Vidal, J.M., Wallace, R., and Wellman, M.P. Toward inquiry-based education through interacting software agents. IEEE Computer 29(5):69-76, May 1996. [special issue on the NSF/ARPA/NASA Digital Library Initiative]
Atkins, Daniel E., Frank, R., Lougee, W., and Willis, K. An Overview of Digital Library Initiatives at the University of Michigan, March 6, 1996.
Wellman, M.P., Birmingham, W.P., and Durfee, E.H. The digital library as community of information agents. IEEE Expert 11(3):10-11, June 1996.
Artemis: Driving Questions on the Web
Comments or questions may be sent to: UMDL.INFO@umich.edu