In this section, we review our progress for the period February 1996 through February 1997. Our work plan described developments divided into three major areas: research, testbed and collection, and evaluation.
Goal 3.a.1 - Design an authorization and security plan (ASP) for UMDL. The plan will be compatible with existing software within UMDL, such as NetBill, and will use existing technologies where possible.
We evaluated our security needs and determined that we did not have the programming resources (people) to do a comprehensive design and implementation. We are using existing technology (such as domain name restriction) to provide basic security to meet our contractual obligations and the needs of our publishers.
Goal 3.a.2 - Integrate the NetBill payment system.
Due to the continuing unavailability of Carnegie Mellon University's (CMU) NetBill software, we were unable to make actual development progress on this goal. Instead, we studied the available NetBill interfaces for purposes of integration planning. Michael Wellman met with Marvin Sirbu at CMU in September 1996 to discuss this, and William Walsh visited CMU in November 1996 to coordinate with the NetBill development team. We also designed a simple internal accounting scheme to use in UMDL in the absence of NetBill. We have the basic hooks to link to NetBill or other payment systems when available.
Goal 3.a.3 - Implement market facilitators (MF), and integrate with a set of UMDL agents.
This goal required that a group of UMDL agents engage the MFs to clear prices for a set of goods.
We made substantial progress on this goal in 1996. The market facilitators (auctions) are now operational within the UMDL prototype, and a complex market-based scenario (the "Service Markets Society" testbed) involving several types of UMDL agents was demonstrated at the Stanford DLI meeting in December 1996. We presented a paper on the UMDL market facilities:
Tracy Mullen and Michael P. Wellman, "Market-based negotiation for digital library services," Second USENIX Workshop on Electronic Commerce, November 1996.
Goal 3.a.4 - Development of a language to allow agents to describe goods, and thus markets.
Substantial progress was made in 1996 towards developing, and using, the UMDL ontology. The content area of the ontology is now developed to the point where are ready to use it to describe a small collection. We have also built an agent that uses the ontology to classify agent services, in a dynamic way that greatly enhances the ability of societies of agents to change and grow.
The Service Classifier Agent (SCA), developed in 1996, is the first UMDL agent based on the ontology. To advertise their services, agents define their capabilities using terminology from the UMDL ontology. In other words, this is the language that is used to describe goods and services in UMDL.
Goal 3.a.5 -Provide a protocol-specification document and realize it within the set of UMDL protocols.
The basic protocol for negotiating through UMDL auctions has been specified and implemented. Additional market-based negotiation protocols will be developed for the Michigan Internet AuctionBot, and transferred to the UMDL auction scheme when appropriate.
Goal 3.a.6 - Develop a comprehensive design for distributing the registry that includes incorporation of the CNRI Handle System.
We have been maintaining and upgrading the registry database system using SYBASE. As part of this effort, we addressed contention and reliabilty needs of UMDL by having multiple copies of the registry being run in parallel. Agents randomly access one of these replicas, and SYBASE keeps their contents consistent.
We looked at the issues involved in a scalable distributed registry, and decided the payoff for distribution was not sufficent to warrant the resources. We also looked at the CNRI Handle System and determined that making use of it would not have a significant impact upon our registry architecture.
Goal 3.a.7 - Investigate the use of other mechanisms for negotiation that build on and extend the Market Facilitators, and the goods and service language and the negotiation protocol it uses.
The challenge of teaming within the UMDL depends on global mechanisms, including auctions to support the discovery of agreement terms, payment systems, and classification systems to allow agents to communicate unambiguously about goods and services. Effective use of these mechanisms requires that individual agents make good decisions about how to employ them. These decisions include decisions about how to decide what offers to make, decisions about what to learn about other agents, and decisions about how to use the services of others effectively.
Agent Modeling Mechanisms
We have used simulation to identify the effects of learning agent models (specifically, models that relate information good prices to expected information quality), to discern the features of an information economy that indicate the usefulness of such learning capabilities. We have also used computational learning theoretic techniques to analyze the complexity of the learning problems such agents face.
Strategic Negotiation During Simultaneous Contracting
We have extended our Markov Model-based tools for computing optimal offers given models of agents interested in goods and services to include reasoning about the effects of other offers being made simultaneously. We could show in simulation how such reasoning can lead an agent to make more competitive offers, such that they are accepted more often and/or are each more profitable than would occur without such strategic reasoning.
Goal 3.a.8 - Revise and update UMDL architecture design document. This includes updates to the protocols, and the availability of "agentware" allowing third parties to develop agents.
We continue to revise and update the document which describes the design and implementation of the UMDL Agent Architecture.
We have developed a publishable (and to be published) detailed description of the Agent Architectural levels and protocols in UMDL.
Goal 3.a.9 - Continued enhancement to the Collection Search and Retrieval Process
Our interoperability protocol that we developed with Stanford University is the means by which our user interface agents (UIAs) send queries to and receive results from collection interface agents (CIAs). For our current set of collections, we determined that the protocol is best used for transferring metadata about documents and not for transferring documents themselves. As such, we have not pursued transporting various document types via the protocol. Instead, we use the protocol to locate documents which are then transferred from the collections server to the users web browser directly using HTTP.
We had initial discussions with Stanford University concerning expanding the interoperability protocol to include services. Due to technical difficulties beyond our control, we have not been able to continue our current level of interoperability with them, and thus have not pursued broader forms of interoperability. (The new release of ILU, Xerox PARC's COBRA implementation does not yet support C++. The new version is not compatible with the old one, and therefore, our digital libraries can no longer communicate.)
Goal 3.a.10 - Web document search and retrieval
We developed a web-book agent in the UMDL Agent Architecutre. Librarians, students, and other participants identify web books to UMDL web-book agent. The web-book agent collapses them into a single web book according to domain name and consolidates them into a single web book. The web-book indexes web-book content so that UMDL users' queries retrieve web-books based on their specific interests.
Goal 3.a.11 Thesaurus development
Switching between the Broad System and Ordering (BSO) and NASA Thesaurus is transparent to users. We are adding the Sears Subject Headings which are geared toward the technical level of high school students to browsing, searching, and switching capablities.
Goal 3.a.12 Collection query transformation
A single query searches FTL, Open Text, and several online catalogs.
Goal 3.a.13 Relevance Feedback
No progress made because the student assigned to this task left to take an academic teaching and research position. This activity will be subsumed by a new evaluation effort on the effectiveness of UMDL Artemis in supporting relevant information retrieval at one of the high school science deployment sites.
Goal 3.a.14 Information gathering interface
We have designed and are implementing an interface to the UMDL which should truly support the query and re-search activity involved in inquiry based science projects for middle and high school students. Named "Artemis" (goddess of the hunt) , it has explicit scaffolding to support students as they learn the science content (earth and space sciences) as well as the skills to carry out an inquiry.
Goal 3.b.2 - Increase size of collection
During 1996-97, the development of testbed collections focused on two areas: incorporation of remote resources and integration of SGML collections. In addition to these emphases, new content contributions were received from Grolier's, Encyclopedia Britanicca, and Cambridge University Press, and government GIS sources adding to the ongoing receipts from our existing publisher partners. Future activity will focus on integration of new format receipts such as GIS.
Given the architectural focus of the project on a 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 (discussed in the ontology and "web book" descriptions in other sections of this report). Thousands of Internet sites were evaluated for relevance to the high school science curriculum in our test sites. One area of particular development has been resources provided by or sponsored by NASA. The topical scope of these collections includes remote sensing of earth from space, our solar system, historical material, probes and satellites, manned exploration, SETI efforts, and deep space research. Data types include text, photographic, animation, numeric data, and visualization material. A number of the resources, such as Online from Jupiter, Windows to the Universe, and the Space Telescope Science Institute, also provide material of use in the classroom.
Continued development on enhancing SGML mechanisms occurred. Grolier's Americana Encyclopedia was added to the testbed and SGML access mechanisms for this and the McGraw Hill Encyclopedia of Science and Technology were developed, focusing on improved performance, user feedback, and display.
Goal 3.b.3 - Add New York Public Library and Stuyvesant High School as test bed sites (perform baseline evaluation at these sites)
The holding pattern continues with New York Public Library, and we are looking at deploying the testbed at Hunter School (rather than Stuyvesant High School which has undergone a change in their administration) since they are a magnet school with a science focus.
Goal 3.b.4 - Deploy release 2 at education test sites
We decided not to deploy the UMDL in the participating high school classrooms this year as the Artemis interface is still being developed. Rather, we chose to use the Internet as a vehicle for exploring how to use online resources as preparation for the students to successfully formulate researchable questions and keyword searches. Thus, we have developed a set of instructional techniques that teachers are using to help students initially develop driving questions (e.g., brainstorm 3 questions before going to the computer lab; go to the computer lab and visit some "hot sites" first to gain a background, and then develop 3 driving questions; develop 3 questions since some questions will not be researchable using online resources). We are in the process of producing a "how to" guide for teachers, where we codify instructional strategies that they might use to better support their students in the various inquiry activities.
We developed and used the following units in the high school classrooms as described in UMDL Teaching and Learning Project.
Goal 3.c.1 - Evaluate precision of search results at testbed sites against different search engines and against linguistic algorithms
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.
Goal 3.c.2 - Evaluate UMDL release 2 at test sites
Over 1,000 middle and high school students in Ann Arbor used our online curriculum units during the year. However, since we were using the Internet and not the UMDL itself, the evaluations have focused on strategies such as teacher development, instructional, student search and inquiry, and student publishing. (The UMDL interface will be used in the classroom starting in April 97.)