Following is a brief summary of objectives and progress of the University of Michigan Digital Library (NSF-UMDL) project from August, 1996 through November, 1996.
During the past quarter, we have been mostly concerned with both demonstrating dynamic resource allocation using economic principles, and creating facilities to make it easier for agents to both describe their capabilities and find other agents.
Dynamic resource allocation in the UMDL architecture is supported, at this point, by the introduction of a new agent, the auction agent, and associated protocols. The auction agent acts as a market for a particular information good or service; each good or service, to some level of abstraction, has its own auction. The auctions have rules to describe their behaviors, such as when the auction is cleared, what constitutes a legal bid, and so forth. Under these rules, the auctions match buyers and sellers, and determines prices for goods and services.
At this point, we have created an prototype set of auction agents that operate under very simple rules (e.g., market clears when the buyer's willingness-to-pay exceeds the seller's price, with the price set by the seller). The prototype "sellers" are task-planning agents (TPA) that sell the ability to process a query; the "buyers" are user-interface agents (UIA) that buy the service sold by the TPAs. The prototype uses a random distribution of queries to simulate buyers at the market, and the TPAs are able to process queries at (user set) rates. Thus, we can vary the distributions and rates and see the distribution of queries to TPAs vary (e.g., "fast" TPAs get more queries than slower ones). The price a TPA sets is based on its load.
As new goods or services are introduced to UMDL, new markets will need to be created for them. This implies that new auctions will be needed. We have prototyped a simple auction-manager agent that decides when to create a new auction.
As mentioned previously, we assume that new agents, therefore new services or goods, can enter UMDL. Yet, in order for agents to find other agents, it is essential that they know the correct "description" of those agents. For example, if a new TPA comes on-line, it must know where the Query auction exists. Since agents may use different names to name themselves, we need some way of searching the name space for similar names. The Service Classifier Agent (SCA) provides this search capability.
The SCA uses a description logic system, LOOM, to construct a directed graph of terms, and then uses the inferencing power of descriptive logics to find matching names. This logic system allows one to define concepts by the conditions under which they hold. The SCA can find the most specific subsuming term that covers another term, for example. Thus, the SCA can help a TPA find query market by mapping a generic description such as "market for high-school level queries" (this is highly stylized English, the real description is much more precise) into a specific term used by a market, such as "auction-K12-query".
The SCA is also integrated into the auction demo. Here, the SCA allows agents to map terms to find specific auctions.
Thus, this quarter has seen the UMDL architecture begin to incorporate economic features, and to provide support for agents dynamic behavior. The prototype will be demonstrated at the upcoming Stanford DLI meeting, December, 1996.
The Advanced User Interface (AUI) group has as its charter to research, design and begin prototype implementation of a visually rich paradigm for the extended information gathering and structuring task. In the past quarter we have continued the task generalization analysis, beginning design activities, and technology explorations.
For task generalization, we have been struggling for an adequate representation of the variety of procedures that our subjects employed in their varied finding/gathering sensemaking activities. Last quarter we came up with the "places model" that seeks to represent how procedures arise as a result of the properties of the available information landscape. This quarter we began a small implementation of the model in PAD++ and found it useful for pointing out some of its representational shortcomings. We also articulated a framework in the form of a process space tree that combines both the part/whole and type/subtype relationships for a space of procedures. We hope next to fill some details from our data and integrate it with the places model. We also have undertaken a more ambitious survey of the seemingly vast amount of relevant literature, are compiling an annotated bibliography, and working towards a more integrated review.
We have compiled a long list of design implications from our task analyses so far, some specific to the interface, others pushing more deeply into the UMDL infrastructure. Based on these implications we have made a first pass at a several high level design sketches and are currently working to integrate these into one or two more serious design which we will pursue in the coming year.
We have been using technology explorations to enrich our understanding of the design space. The explorations have taken the form of small "microprototypes", and have been built on PAD++, a infinitely zooming worksurface, scripted with TCL/TK to provide fast prototyping. Among the new micro-prototypes we have tried is a new "split-zooming" technique that allows zooming-in to split smoothly into separate portals so that multiple regions of interest can be kept in view. We expect this to be important for navigating around the information visualizations and multiple task activities which will be in the final interface. (This work is done in close association with the PAD++ ARPA grant). We have also started to look at ways to evaluate of our micro-prototypes, difficult since they are merely rough implementations of fragments of functionality, and not ready to put directly in front of users. We have compiled a list of possible techniques and hope to try some on our query history micro-prototype. Towards that end we have also been working to hook up our prototype front ends to the back end search functionality of the existing UMDL agent architecture.
To respond to the site-visitors' requests that we more fully document our process we have been keeping extensive on-line records of our activities, including meeting notes and an extended web document. The AUI group has also submitted a Supplement application to help support and expand its activities for the remainder of the project.
Effort continues to add additional locally hosted content. Test data from Cambridge University Press was received (although the data appear problematic and may need significant work to deploy) and licensing is near final with Encyclopedia Britannica. A number of solicitations for content are also active, although slow to reach closure.
Given the architectural focus of the project on a highly heterogeneous, distributed library, a priority has been put on expanding the number and diversity of remote collections and enhancing the registration and metadata descriptive process.
Considerable energy this quarter has been directed toward several related tasks surrounding the identification and representation of collections in the UMDL. These tasks include an update and expansion of the UMDL ontology, a revision of the UMDL metadata set, enhancements to the collection registry process, and the identification and registration of new sites.
Work on the UMDL ontology has adopted a structure initially proposed by the International Federation of Library Associations for published information and enhanced through added terms and definitions and more formally described facets of information (for examples of work see http://www-personal.engin.umich.edu/~peterw/Ontology/ontology.html).
Concurrently, applications are being designed to take advantage of the evolving ontology. For example, it has been used as the basis of a Service Classifier Agent. The Conspectus group is currently also testing the ontology by creating a small set (500-1,000 items) of material which will be imported from USMARC into a knowledge base using the ontology, allowing testing of the logic and functionality of the ontology.
A revised set of metadata drawn from the ontology has been used to register new collection resources. The resource registration interface has been redesigned to include editing
capability, defaults, and confirmatory information to the information provider. Coupled with the web-book mechanisms developed last spring, the new registration process allows groups of pages under a given home page to be registered together under a uniform title as a related entity, and indexed together. Heuristic rules were also developed which define the boundaries of each web-book collection, index, and eliminate duplication in the registration process. The new registry structure allows for better presentation of results on the user interface side, navigation support for registrars, and the possibility of automated inclusion of new material in a registered site (see http://polyphemus.engin.umich.edu).
Concurrent with the development of the registry process and tools, thousands of sites related to Astronomy, Water Ecology, Weather, and Geology (geared towards middle and high school audiences) were evaluated for a match with the user communities' needs and a subset was ultimately selected and registered. This selection/registration process, coupled with the web-book indexing tools have significantly increased the collections represented such that some 7,000 content sites have been identified for registration (an increase from several hundred previously).
The PAT Collection Interface Agent has matured into supporting the human selection of Web based materials for UMDL. The Collections Group is using the web-book concept to gather science materials from the Web into a searchable collection maintained and serviced by the PAT CIA. To achieve the goal of supporting the web-book collection a number of changes have been made to the PAT CIA. First, the PAT CIA has been converted over to the common system architecture used by FTL, Z39.50, and other CIAs to further bring it into a production system. Next, the web-books placed in the collections are gathered from the human selected web-books that are registered in the conspectus. Finally, because of the improved quality description of each web-book, the indexed web-books are enhanced with added entry points.
The information records that describe each web-book at the conspectus level are used to collect the Web pages when creating the collection. This leverages the expertise of librarians when building collection, by letting librarians do the selection after a Web-book browser tries an initial cut at a web-book. This maintains the high quality in a collection.
You would lower the precision of materials if you totally automate the selection of Web pages and leave out any human oversight.
By using the information from the conspectus records of each web-book, the PAT CIA is adding high quality access points to improve searching the web-book materials. Each Web-book will have the Title, Author, Subject Heading, Description, Format, Conceptual-Level all added with special HTML tags in the head page of the web-book. By using special tags, the user can then focus their query to a very precise attribute that will occur for EACH web-book. No longer does the collection rely solely on the content of a Web page for words to search on. The PAT CIA will translate the UMDL query attributes into the special HTML tags to make them available to all users.
During the past quarter the Education Group engaged in the following activities:
Curriculum Development: We are preparing units for Geology, Weather, Astronomy, and Water Resources. These units will be used in both middle schools (in our NSF supported Middle Years Digital Library Project) and in the participating high schools.
Collection Development: In preparation for the availability of Artemis (the new UMDL interface) in March 1997, we have collecting and cataloging web sites that pertain to the curriculum units we are developing. We are attempting to gather 200 sites for each of the topics.
Artemis Roll-Out: We have planned the phasing in of Artemis, the new UMDL interface that was expressly designed to support inquiry-based science learning. We carried out an Alpha-1 test, in the first phase of the roll-out.
Artemis Evaluation Study: We are in the midst of designing the study that will evaluate the effectiveness of Artemis in the classroom.
Over the last quarter, we have continued our development and integration of commerce mechanisms within the UMDL architecture. We designed protocols for automatic generation of markets for information goods and services, to be demonstrated at the December DLI meeting. The auction manager determines which markets are applicable for which agents, and creates new service descriptions and markets when appropriate. Several additional agent types have been enhanced to participate in market-based negotiation.
There have been two major thrusts in our testbed construction activities. One thrust has been to improve robustness and increase functionality to support the ongoing research in agent building, commerce, and search and retrieval. The other thrust as been to complete the functionality required to deploy UMDL for use in the Ann Arbor Public Schools to support our education research.
In the former, we have improved the agent infrastructure (Agentware) to include greater support for building multithreaded agents. This will allow us to use more flexible programming models and to improve overall throughput and performance by increasing concurrency in agents. We have also added functions to support instrumentation of agents during performance and behavior experiments. It is now easier to trace the sequence and timing of agent activities.
We have continued our development of a suite of tools for building and testing agents. Our latest addition is a tool which acts as a client to a user-selected agent. It allows the developer to exercise and test the agent by sending it messages and displaying the agent's responses. The developer has complete control over the content, timing, and order of messages sent to the agent.
Our educational deployment activities have concentrated on continuing development of Artemis, the Java-based UMDL user interface, and the user interface agent (UIA) which provides the link between Artemis and UMDL. A number of functions have been added to Artemis. We have changed the list of broad topics to only include those in the area of Astronomy and Astrophysics. These are the areas where Artemis will first be used in the schools. Search results are now displayed asynchronously as they are retrieved. The user does not need to wait until the search has been completed by all collections to view results. We have implemented functions to create and view Driving Question Bins and have added the ability to drag individual documents or entire result sets from the search result list to a Driving Question Bin. Items in Driving Question Bins are persistent across sessions. By allowing users to drag items there, view them, and save them, we give them the capability to organize items into structures that are meaningful to the user.
To support these function provided by Artemis, over the last six months we have thoroughly redesigned the UIA. We have used an object-oriented approach to couple the objects viewed and manipulated with Artemis with the CORBA objects used by the collection search and retrieval protocol. The UIA objects provide access to the data and protocols of the CORBA objects while adding the persistence and ability to organize items which is required by Artemis.
Below is a brief list of some of the meetings and visitors with which NSF-UMDL personnel have been involved during the past months.
9 Wendy Lougee presented 2 papers at TICER International Summer Institute, Tilburg, The Netherlands
29 Wendy Lougee: Presentation for Alice Lloyd Hall Scholars Program
3 Michigan Library Association Summit on Public Policy and Information Access, Keynote Speaker, (Daniel Atkins)
5 Alfred Scheidegger, Urs Spiess, & Oliver Strohm , Eidgenossische Technische Hochschule, (Swiss Federal Institute of Technology) (Wendy Lougee)
9 Visitors from Japanese Universities (Daniel Atkins, Wendy Lougee)
10-11 ALA visitors, Prudence Dayrymple & Elizabeth Martinez (Daniel Atkins)
12 Carnegie-Mellon, Marvin Sirbu, NetBill in UMDL (Michael Wellman)
18-19 National Digital Library Federation Task Force, Washington, DC (Wendy Lougee)
20 Conrad Guetter, Journals Director, Cambridge University Press (Wendy Lougee)
24 Roland Dietz, Vice President, Elsevier Science (Wendy Lougee)
26 Jane White, Sales Representative, Wiley & Sons (Wendy Lougee)
26 Presentation to Warner-Lambert staff (Wendy Lougee)
26 NSF meeting with George Lee (Daniel Atkins)
28 Digital Library Workshop for President's Weekend (Wendy Lougee)
30 Keynote address at Michigan Library Association meeting (Daniel Atkins)
2 Michael Simmons, Santa Fe Institute (Daniel Atkins)
3 Mario Gonzalez, Vice Chancellor, University of Texas (Wendy Lougee)
3 Columbia University External Advistory Board meeting (Daniel Atkins)
5 Plymouth-Canton School District, Technology Day (Joe Krajcik)
6-7 DARPA/ITO PI meeting (Michael Wellman)
9 Ann Wolpert, Director of Libraries, MIT (Daniel Atkins, Wendy Lougee)
11 The Community of Learner's Project, an NSF sponsored NIE project in New Brunswick, Maine (Elliot Soloway)
12 North High School, Omaha, NE, workshop on the use of the UMDL (Joe Krajcik)
15 Iowa's Computing Using Educators State Conference (Elliot Soloway)
17 John Evans, formerly of CSPAN (Wendy Lougee)
23 George Needham, Jeff Johnson, Library of Michigan (Wendy Lougee)
23 ASIS Conference, Baltimore, Maryland presentation on Planning Curriculum for the Information Professionals of the 21st Century (Daniel Atkins)
24 Howard Hughes Medical Institute, K-12 Science Projects (Elliot Soloway)
24 NSA, Columbia, Maryland, presentation on Socio-Technical Prospects and Challenges for Supporting Distance-Independent Knowledge Work (Daniel Atkins)
25 CAMLS Conference, Cleveland, Ohio, presentation on Major Changes in Library Profession: Education for Pre and Post Graduates (Daniel Atkins)
28 Roland Dietz,Vice President, Elsevier Science (Wendy Lougee)
1 John Seely Brown, Xerox Park (Daniel Atkins)
4 Doug Smith , Dorothy Heyart, VPs, Walsh Business College (Wendy Lougee)
6 Presentation at ECOINFORMA Conference (Wendy Lougee)
8 ERCIM Board of Directors meeting speech, Prague (Daniel Atkins)
11 Technical University at Darmstadt, Germany lecture on Knowledge Work Environments of the Future: Experiences and Projects in the USA (Daniel Atkins)
15 Presentation to representatives of the Society for Automotive Engineers at Ford (Wendy Lougee)
18 Charles Ellis, President, John Wiley Publishing (Wendy Lougee)
18 Edwards Brothers board meeting (Daniel Atkins, Wendy Lougee)
18-21 USENIX E-Commerce workshop (Tracey Mullen)
19 Paul Beavers, Assistant Director, Purdy/Kresge Library, Wayne State University (Wendy Lougee)
19-21 Carnegie-Mellon University, payment mechanisms in UMDL (W. Walsh)
20 Wendy Lougee presented Keynote address at Michigan Library Consortium Conference
20 UMDL presentation at the Technology Day in Lansing, MI (David Lyons)
21 Greg Badger, Bowker International/Reed Reference (Wendy Lougee)
November 27, 1996
I certify that to the best of my knowledge (1) the statements herein (excluding scientific hypotheses and scientific options) are true and complete, and (2) the text and graphics in this report as well as any accompanying publications or other documents, unless otherwise indicated, are the original work of the signatories or individuals working under their supervision. I understand that the willful provision of false information or concealing a material fact in this report or any other communication submitted to NSF is a criminal offense (U.S. Code, Title 18, Section 1011).
Project Director Signature _____________________________________________
Daniel E. Atkins, Dean
School of Information
University of Michigan