The main progress in IPE over the preceding period is improvement in the market infrastructure underlying the SMS research testbed. In particular, we have extended the auction manager to provide more flexible control over when and which markets are created by agents seeking to buy or sell library services. In addition, we have increased the variety of types of auctions available in the system, by adopting the configurable auction framework provided by the Michigan Internet AuctionBot (see http://auction.eecs.umich.edu/).
The Service Markets Society brought together research in markets, ontologies, and strategic behavior. Markets are used to balance queries among a set of Task Planning Agents (TPAs). The Service Classifier Agent (SCA) uses the evolving UMDL ontology to uniquely name classes of agents which perform the same services.
The SMS currently includes six types of agents (Task Planner, Service Classifier, Auction Manager, Auctioneer, User Interface and Registry), with potentially dozens of active instances. During this period, work on expanding the capabilities of the existing Web Agent, Agent Interpreter, and Stochastic Agent for Multiagent Contracts continued.
The SMS team has started defining a specification for a new type of UMDL agent called a ìPreference Agentî. This agent will enable each user to have a personalized ordering of results based on his/her preferences. It will also enable results from multiple collections to be scored and combined. To facilitate the Preference Agent, we have focused on trying to improve document level metadata from our existing collections.
Information on the development of Preference Agents may be found at:
http://www-personal.umich.edu/~compuman/UMDL/cia-metadata.html and at
http://www-personal.umich.edu/~compuman/UMDL/pref_spec.html
Work continued on the UMDL agent-building toolkit; a set of tools for simplifying the task of building agents that are compliant with the UMDL Agent Architecture. This toolkit allows users to design agents based on the protocols that they engage in to achieve goals, while abstracting the complex code needed for communication. These protocols can be created specifically for new types of agents, or can be combined with previously designed protocols supplied in a protocol library supplied with the toolkit. The toolkit will contain protocols already used by UMDL, as well as related agent systems, where possible.
These protocols define the type, sequence and content of messages exchanged among a set of agents. Once the agent is specified in terms of the protocols it uses, skeleton code will be generated for the agent that implements the communication necessary for the agent to interact with other agents. Users will then fill in the necessary decision-making capabilities to make a fully functioning agent.
Agent-building toolkit development involves examining various agent programming architectures to see what architecture is closest to our needs. Our current interpreter is based on MYWORLD which was created by Mike Wooldridge.
We have a Prolog prototype that demonstrates the interpreter flow and the logic of an agent program. Agent programs are written in Prolog. Currently, we are examining how to interact this interpreter with current C/C++ code.
We have also experimented with Prolog and C/C++ interaction, so that we can implement parts of the interpreter in Prolog. We have also experimented with Lisp. We are examining if this is a better alternative to Prolog.
We have continued the development of concept definitions in the ontology, and also began to test our work by using the ontology to describe a collection. This first application of the content area of the ontology demonstrated the feasibility of developing ontology-based catalogs various intellectual works, and the significant benefits for powerful queries that result.
We have written programs to view and manipulate US MARC data; developing a mapping from the MARC fields to our ontology; an initial design for the software to do the conversion; and further refinement and extension of the ontologies in several areas.
We have further elaborated and defined metadata elements drawn from the ontology, with at least one more iteration of metadata describing WORK and creation of metadata attributes and domains describing services.
The Advanced User Interface group has continued its design efforts, exploring a number of desired properties that arose from our earlier observational research. We have focused on developing two prototypes.
The first prototype is pushing hard on queriability, navigability and history in a multi-scale zooming metaphor with some new interface widgets and interaction styles. The prototype is being implemented in the infinite zooming workspace, PAD++, and also involves the development of our own small, vector retrieval engine for the user's local workspace. One aspect of particular interest is the fact that the interface has no special query window. Anything on the worksurface can be used as a query against any set of other items on the surface. For example to submit a query of the standard sort, one just types the query text anywhere on the worksurface (e.g., near where you are currently working). This new text object is "thrown" against some designated set of other objects on the surface (e.g., structured collections of other items) and the items similar to the query are highlighted. The user can then browse through the results in the context of their source structure. The result is a tighter integration of the query operations with both the context from which the query sprang and the context in which the results are situated, needs that came up in our interviews. Another result is several interesting synergies of query and navigation.
The second design prototype is pushing more explicitly on the shared, social possibilities of cyber information access for the DL. One inspiration is the insight that while physical libraries require relative quiet to prevent physically collocated but logically unrelated researchers from disturbing each other, there is no such constraint in an electronic library ("No need for silence in the digital library"). We created a first prototype cyber library space where information droids, hooked into the current UMDL and an additional image database, answer questions from a group human participants, and they can all interact with each other and the query results in various ways. The prototype has been built on a new substrate called CCR, similar to a computationally rich, distributed peer-to-peer MOO, being developed by David Ackley on a DARPA grant.
Our plan for the coming period is to further extend the auction manager and the range of auction services available. We will also be integrating the market functionality in the SMS testbed with the operational system, so that all agents can interact through market interfaces. Our next milestone in this regard is to demonstrate a scenario of agent team-formation via market-based negotiation protocols.
In the second half of 1997, we plan to scale up the SMS to conceivably include thousands of agents. This simulation will provide a testbed for demonstrating and testing many facets of our research. These may include:
Preference Agent Development: By the end of October, we plan to have an operational prototype Preference Agent. This prototype will be integrated into Artemis (for deployment), and will add the capability of document ordering and scoring. In addition, we plan on improving document level metadata to the HotBot CIA, as well as other agents. By the end of December, a second prototype will be developed based on experience with the first one.
Agent-Building Toolkit: By the end of August, we hope to have a working prototype of the interpreter. We will create several agents using the new language that will imitate the UMDL agents. We also will have a translator that will translate the language into UM-PRS code. By the end of September, we hope to have the language and interpreter worked out so that the agents can be integrated into the UMDL system without affecting the current set of agents. By the end of the current calendar year, we hope to make the interpreter abstract so the interpreter can be used with the various agent systems.
For the next period, we plan to extend the ontology into describing services that would exist in the digital library. This will enable us to more fully integrate SMS into the digital library.
Continue development of prototype advanced interfaces. Based on the observation that silence is not necessary in the digital library, we will explore collaborative group information seeking using a spatial metaphor populated by people and ìinfobotsî in the CCR distributed computation environment. Pushing a novel combination of visualization, navigation and querying, we will continue to develop our multi-scale information search space, implemented in the infinitely zooming workspace, PAD++.
Since the last semi-annual report, the Webbook collection interface agent has developed into a permanent collection in the UMDL. There have been improvements to the Webbook collection process to improve accuracy and performance. Along with collection process improvement, there has been an improvement to the searching for data in a Webbook and the display of a selected Webbook that was found when searching the collection.
The collection maintenance process has gone through a large-scale test by building a significant collection. The system has built a collection with over 400 Webbooks, over 4000 pages in the Webbooks and over 100-MB of data from the World Wide Web. Gathering pages off the WWW needed fine-tuning to take into account slow connection to servers and time-outs and to allow the process to be run without too much interference with other processes on the same server. The intent is to be thorough and gather the complete list of sites since each site is valuable to the collection.
The collection process has been further fined tuned by giving more control to the librarians when defining a Webbook. The previous system only allowed a librarian to define the start or head page of a Webbook. After setting that page, the automated Webbook collecting tool searches and collects the underlying pages that made up that specific Webbook. Now, that process can be done under the control of the librarian. The librarian first gets a preliminary list of pages for the Webbook from our automated tool. The librarian can then pick and choose the specific pages to use for the Webbook. Once the librarian is satisfied with the list, it is passed on to the collection maintenance process. That process will not use the automated collection tool, but will collect only the list chosen by the librarian. So now a collection development person can let the system pick the pages for a Webbook or they can pick the pages themselves after an initial set is automatically chosen for them
Changes have been made to improve the descriptive information produced by the librarians when they do collection development for a Webbook. Those improvements further define the type of data contained in each Webbook which can improve searches. The improvements have been to further distinguish between forms of the Webbook by splitting the description of form into published-form and genre-form.
The search and retrieval part of the Webbook collection interface has been improved and display of records integrated with the new template form of display. The collection interface agent now formats records by sending the user interface agent an HTML template, which fills in fields for each Z39.50 field, needed on the display. This new display has allowed the Webbook CIA to show the basic information about a Webbook, its title, author, summary and URL and all of the hits found any of the pages within the Webbook.
The Webbook collection interface agent can now process searches on specific fields within each Webbook by using the USE search attribute. This allows searchers to further refine their searches to pinpoint parts of the Webbook structure when doing a search
The testbed has continued to grow through the addition of new commercial content and efforts to identify relevant resources freely available on the Internet. A significant enhancement resulted from a shift from a local implementation of the UMI journal collection to integration of the publisher host (increasing the titles from 50 to nearly 200 and articles to over a half million). New editions and supplementary images of reference works were also added, increasing and updating this core curricular content.
Enhancements to the collection registry capability allowed stepped up inclusion of Internet resources. Capabilities to "harvest" indexes on the Internet for suitable sites and autoregister and manage these resources were created. Over 400 sites (over 3500 homepages) are now registered.
Testbed development of the production system in the past six months has been focused on evaluating and testing UMDL in real use. Five classes of students each used UMDL during four daylong units. Two units, Astronomy and Geology, were studied. Based on observations and feedback from this use we have modified UMDL to improve performance and reliability. We have also modified the user interface, Artemis, to provide better support for functions that were deemed the most useful and to eliminate some features that were used less frequently.
The Artemis interface provides users with a persistent, individual workspace to support student engagement in science high school inquiry. The workspace is preserved between sessions, and the same workspace may be accessed from school, home, or library machines. Implemented in Java, it is platform independent.
A target set of functions has been identified for the full deployment. Updates will be released as needed, most probably on a monthly basis. We will also be concentrating on identifying new services that can be incorporated into UMDL. We will both provide access to existing services on the World Wide Web and implement new services. In addition, we will continue to expand the access to collections that the University of Michigan has licensed, but which are currently not accessible via UMDL.
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 have some baselines from this yearís evaluation effort that we may be able to use for comparative purposes. We also plan to test scaffolding techniques to determine their usefulness in ranking large retrievals from very large databases.
Sample questions which the evaluation form will gather data include:
Work will commence in fall to involve the New York Public Library in identification and construction of resources suitable for public library users. This extension of registration process to third parties will further test our registration tools.
The University of Michigan Library's partnership with Elsevier Science has created a host service of all 1100 Elsevier journals, with an associated pricing research project. These journals will be incorporated in the testbed in the future. In addition, 15 gigabytes of census and map data from Tiger Inc., and another gigabyte of basic world map data from ArcView will be added with basic display and access functionality. Future additions to the site will include UNIX based tools and direct web based access on all platforms to GIS data. The primary audience of this site will be University users.
Joseph Hardin, formerly associate director of NCSA for software development has joined the staff of the University of Michigan School of Information effective August 11, 1997. Joseph is now director of the Systems Develop and Operations Group for the School. Included in his group are all of the professional research programmers supporting the UMDL and other experimental system building activities. A major part of Joseph's position is to further enhance the environment for successfully carrying out research involving experimental system building.