|The University Libraries of the University of Houston created an experimental Intelligent Reference Information System (IRIS) over a two-year period. A ten-workstation CD-ROM LAN was implemented that provided access to nineteen citation, full-text, graphic, and numeric databases. An expert system, Reference Expert, was developed to assist users in selecting appropriate printed and electronic reference sources. This expert system was made available on both network and stand-alone workstations. Three research studies were conducted.|
The principal goals of the IRIS Project were to: (1) implement a CD-ROM LAN that would provide access to citation, full-text, graphic, and numeric databases; and (2) develop an expert system that would recommend appropriate CD-ROM and print reference sources.
Three research studies were conducted: (1) a CD-ROM LAN performance benchmark; (2) a survey of user perceptions of the CD-ROM LAN; and (3) a survey of user perceptions of the expert system.
The IRIS project was intended to assist all members of the university community, plus the numerous Houston citizens who use the services of the University of Houston Libraries. This user population is ethnically diverse and multilingual.
The IRIS Project evolved from two earlier projects: (1) the Intelligent Reference Systems project, which developed an expert system for indexes and abstracts (Index Expert)1, 2; and (2) the Electronic Publications Center project, which established a CD-ROM service that employed stand-alone workstations.3
EDUCOM's Educational Uses of Information Technology (EUIT) Program acknowledged the accomplishments of the IRIS Project when it named the project one of its Joe Wyatt Challenge Success Stories. The Joe Wyatt Challenge was intended to identify 100 successful applications of information technology in U.S. and Canadian colleges and universities; 101 projects were actually chosen.
The IRIS Project Director was Robin N. Downes, director of the University Libraries. Reporting to Downes, the Project Management Group supervised the efforts of the Electronic Publications Instruction Group (bibliographic instruction and user documentation), the Knowledge Engineering Group (expert system development), and the Research and Evaluation Group (CD-ROM LAN performance benchmark and user studies). The final project structure was as follows:
Since CD-ROM vendors were uncertain about how to price network licenses to CD-ROM products, negotiations with vendors were lengthy and there was little similarity in the obtained license agreements.4 Initially, vendors restricted access to their products in various ways, such as the number of overall network workstations, the number of simultaneous users per database, and the workstation location. When the licenses were renewed, many vendors focused on the number of simultaneous users as the primary way of contractually limiting database access.
The nineteen networked CD-ROM databases used in the IRIS Project were:
The IRIS Project designed and installed a ten-workstation CD-ROM LAN. The workstations were 16 MHz 80386SX microcomputers with 1 MB of RAM,4O MB hard discs, and EGA monitors. The project used an IBM Token-Ring LAN that ran under Novell Advanced NetWare 2.15 Revision B. Two Meridian Data CD Net 314 CD-ROM servers were employed. A 20 MHz 80386 server was used to support NetWare and centrally provided software. The Meridian servers and the network workstations used the Meridian CD Net, NetBIOS, and NetWare DOS Client programs. The Microsoft CD-ROM Extensions software was used on workstations. The Saber LAN Administration Pack was used to provide menu-driven access to CD-ROM resources, log CD-ROM sessions in a dBASE-compatible file, control access to CD-ROM resources to comply with license agreements, and provide enhanced network security. From eight public workstations, library patrons were easily able to access any desired CD-ROM database. Library staff used two additional workstations: (1) the Information Services desk workstation was employed for ready reference searching; and (2) the computer room workstation was used to manage the CD-ROM LAN. During most of the project period, there were also four stand-alone CD-ROM workstations in the Electronic Publications Center that provided access to five checkout databases (there were three workstations in use after Reference Expert was implemented).
The CD-ROM LAN implementation process was challenging.5 The purchase of "plug-compatible" Token-Ring boards resulted in compatibility problems with the Meridian CD-ROM server software that needed to be resolved. Some network cables were defective and had to be replaced. It was apparent that few CD-ROM vendors had designed their searching software to operate on a network, and the process of getting a CD-ROM database up and running on the network was often far more complex than it should have been. Some CD-ROM searching software would run from the Novell NetWare server; some would not. Workstation memory was another problem area; the AboveLAN software was required to free up conventional memory in order to run some CD-ROM searching software. (At a later date, migration to MS-DOS 5.0 made it unnecessary to continue using the AboveLAN program.)
The CD-ROM LAN was made public in August 1990. It has been heavily used by the patrons of the University of Houston Libraries, and it has significantly increased user access to electronic information. In the first year of the grant (October 1989 to September 1990), CD-ROM databases were used 25,264 times (7,628 network uses in the two months that the network was available). In the second year of the grant (October 1990 to September 1991), CD-ROM use skyrocketed to 77,031 database uses (68,784 network uses), which reflects the fact that the network was available during the entire year. The highest number of CD-ROM database uses in a single month of the grant period occurred in April 1991, with 9,678 uses (8,581 network uses).
Effective management of the IRIS CD-ROM LAN required several policy decisions.6 In order to provide equitable access to the popular CD-ROM LAN, workstations were scheduled in one-half-hour blocks during peak use periods. To provide the widest breadth of resources, very few CD-ROM back files were mounted on the CD-ROM servers; the majority of back files were made available on a checkout basis. Staffing was significantly increased in the Electronic Publications Center to provide more user assistance.
To support the effective use of the IRIS CD-ROM network, the University Libraries developed new user documentation and expanded its instructional efforts. It created a Quick Reference Card for each CD-ROM database that provided brief instructions for using the database.7 The Quick Reference Card was available as a handout. It was also included in a three-ring notebook that provided users with more in-depth information about the CD-ROM database, including an Advanced Search Tips guide and, for citation databases, a list of indexed journals. Staff also modified existing publications, course-related workshops, and ongoing library tours to include coverage of IRIS resources. Special hour-long classes about generic CD-ROM searching techniques were initiated.
The CD-ROM LAN has been highly reliable. There is a scheduled weekly maintenance period for the installation of new CD-ROM releases, hardware and software upgrades, minor workstation repairs, and other purposes.
Much of the expert system design process focused on questions about what user, information need, and reference resource characteristics were important and how these characteristics were related to each other. The Knowledge Engineering Group (KEG) identified numerous potentially useful characteristics, but determining how they related to each other was very difficult. Those relationships that were deemed to be important had to be embodied in a menu-driven user interface.
After lengthy analysis of the reference process, KEG determined that the central variables in Reference Expert would be three reference resource characteristics: content type (e.g., addresses, citations, and definitions), format (i.e., CD-ROM or print), and subject coverage.10 It was a significant breakthrough when KEG decided that the traditional category that a reference work was placed in (e.g., dictionary, directory, or index) was not important. Reference works in the same category could be quite different (e.g., an engineering handbook and a psychology handbook). The key to describing reference works effectively was to identify the kinds of information that they contained--their content types. This simple idea was a powerful tool for classifying specific reference works.
The Knowledge Engineering Group developed a frame-based knowledge representation scheme to describe reference materials. Frames are a compact and flexible way of organizing knowledge in a hierarchical structure.11
To build the knowledge base, KEG selected 340 printed and CD-ROM reference sources, and it coded these reference sources using the knowledge representation scheme. This effort focused on the most heavily used reference sources in the collection. KEG recorded title and location information for each reference source, and the committee assigned appropriate subject headings and content types to the work. Comments about the proper use of the reference source were added as needed, and information about the coverage (i.e., selective or comprehensive) of indexes was optionally included for those sources. KEG recorded the relationship between each subject heading and its content types, which required that the committee identify what content types had been assigned to the sources classified under the subject heading. KEG also added descriptions for major subject headings to the knowledge base. The resultant knowledge base was fairly large, containing over 230,000 bytes of information.
The subject heading scheme used in Reference Expert was adopted from an earlier one used in the Index Expert system. The content type scheme was created from scratch. Both were refined in an iterative fashion, with changes being made as new sources were examined.
PDC Prolog, a logic programming language, proved be a good tool for developing Reference Expert's inference engine. An inference engine is the expert system component that emulates the human reasoning process. The program was designed so that the entire knowledge base was loaded into memory from a disc file. The inference engine then used the information in the knowledge base to recommend reference sources. The knowledge base was near its maximum size for network workstations, which had less free memory than stand-alone workstations.
Unfortunately, PDC Prolog was a difficult tool to use for interface design tasks that would be relatively simple with a procedural language like C. The backtracking mechanism used in PDC Prolog, which searches the knowledge base for additional information when the inference engine reaches a dead end in its reasoning process, gives the language great power and flexibility; however, it also makes it difficult to program the parts of the system that require a predictable sequence of activity. Backtracking can be selectively disabled, and this was done, but this is a rather arcane art. A considerable amount of energy was devoted to making the user interface visually attractive and easy to use.
Given the significant differences between logic and procedural programming languages, a novice logic language programmer faces unknown territory, where his or her prior procedural language programming experience may be more of a hindrance than a help. The fact that the University Libraries had previously developed the Index Expert system using both Turbo and PDC Prolog made the programming effort go more quickly than it would have otherwise.
Although PDC Prolog made more effective use of conventional memory (first 640 KB of memory) and ran faster than the expert system shells that were employed in the project, the size of the knowledge base severely degraded system performance until a method of reducing system load was devised. Since this solution required that portions of the knowledge base be deleted from memory during program execution, the knowledge base had to be reloaded each time the program ran. System performance was now speedy, but there was a startup delay each time the system began a new session. It was decided that this was an acceptable tradeoff. It would have been possible to store the knowledge base in a B+ tree disc file; however, this would have made the program considerably more complex. Subsequent testing on an 33 MHz 80386 workstation revealed that this platform provided very good system performance.
Many expert systems developed by libraries use low-cost expert system shells, and they often have a fairly limited scope. For example, a recent survey of Association of Research Libraries members found that four out of the six identified expert system development projects were using expert system shells.12 Expert system shells can be effective for small-scale systems, but may not be adequate for larger, more complex systems.13 The experience of the IRIS Project shows that expert system development with a logic programming language also can push the limits of affordable microcomputer technology.
Originally, we had intended to connect users to recommended CD-ROM LAN databases from within the expert system. It was decided that, given the ease of CD-ROM access from the LAN menu system, this linkage was unnecessary. Since many libraries who might be interested in using Reference Expert may not have CD-ROM LANs, omitting this feature also improved the transportability of the expert system.
Reference Expert was made public in June 1991. The system was available on the ten CD-ROM LAN workstations, on three stand-alone CD-ROM workstations, and on a dedicated workstation at the entrance of the library. Preliminary data indicate that it will be a popular service. From the time the system became public to the close of the grant period (June 1991 to Sept. 1991), Reference Expert was used 3,571 times. There was a steady increase in use of the system: 229 uses in June, 656 uses in July, 937 uses in August, and 1,749 uses in September. (These figures exclude use on a staff workstation used to manage the network.)
The system works as follows. After viewing an introductory screen, the user is presented with a subject menu. Each subject is described in more detail in a window at the bottom of the screen. If the user selects a subject that has lower-level headings in the subject hierarchy, the user can either choose the current subject heading or pick a lower-level heading. The user is shown the lower-level subject headings in a window at the bottom of the screen. After a subject is selected, the system determines whether there are content types associated with the subject (e.g., addresses, brief biographies, or definitions), and if so, it displays a content type menu. Content types may be qualified by language, such as "Translations of words (English, French)," location, such as "Addresses (Houston)," and other criteria. Once the user selects a content type, the system determines if both print and CD-ROM resources exist for the chosen subject and content type, and if so, it displays a format selection menu. Finally, the system retrieves reference sources that match the selected subject, content type, and format criteria. After reading the list of sources, the user can print it. The current date, selected subject, and selected content type are automatically recorded in an ASCII log file, which is designed so that it can be easily imported into a dBASE-compatible system.
Two versions of the system are used in the University Libraries: (1) the version used on LAN workstations that executes once, returning the user to the LAN menu; and (2) the version used on the stand-alone workstation that automatically reloads after executing.
Reference Expert is available at no charge.14 Recipients are licensed to use the program for noncommercial, educational purposes. As of July 1992, over 400 copies of the program have been distributed to libraries, library schools, computer centers, and other users. A 16 MHz 80386SX computer with an EGA or VGA monitor, 1 MB of RAM, and a hard disc is the minimum recommended hardware configuration needed to run the system.
The performance benchmark showed that response time increased substantially as the number of simultaneous users of a CD-ROM LAN database increased. For example, the average increase in response time between one user and nine users for three of the CD-ROM databases (ERIC, Humanities Index, and PsycLIT) was 59.09 seconds. The benchmark also revealed that the degree of performance degradation under load varied considerably by CD-ROM product With nine simultaneous users, there was a 65.08 second difference between the fastest and slowest response time for the three previously mentioned databases. The results of the performance benchmark reflect the specific hardware, software, and CD-ROM databases used by the IRIS Project and the particular testing methodology employed. Given the benchmark results, we currently plan to limit access to a maximum of ten simultaneous users per CD-ROM database.
The CD-ROM LAN survey indicated that the majority of users reacted very favorably to CD-ROM databases, saying that they found information more quickly (89.4%) and more easily (84.2%) than in printed sources. Most users (85.1%) believed that they found more helpful information in CD-ROM databases than in print sources. An overwhelming majority of users (97.5%) agreed that it was valuable having CD-ROM databases in the library.
The Reference Expert survey revealed that the majority of users preferred to use Reference Expert than to use printed guides (62.9%) or to ask library staff for assistance (56.3%). Most users said that they would consult the sources recommended by Reference Expert (72.5%) and use the system in the future to find reference sources (74%).
Given the success of the IRIS Project, the University Libraries are planning to expand the CD-ROM network significantly, increasing both the number of workstations and the number of networked CD-ROM databases.
The University Libraries have established the Reference Expert Task Force to continue the development of the expert system. This group will design and program a prototype of the next version of the system using Visual Basic in a Microsoft Windows environment. The University Libraries are interested in adding significant depth to the decision making process in Reference Expert by taking into account more reference resource characteristics and adding both user and information need characteristics. The University Libraries are also interested in exploring the use of an object-oriented programming language like C++ for producing the production version of the new Reference Expert system. This should reduce the complexity of interface design and increase the overall capacity and performance of the system.
2. Charles W Bailey, Jr., and others, "The Index Expert System: A Knowledge-Based System to Assist Users in Index Selection," Reference Services Review 17, no. 4: 19-28 (1989).
3. Charles W. Bailey, Jr., and Kathleen Gunning, "The Intelligent Reference Information System," CD-ROM Librarian 5: 14,16 (Sept. 1990).
4. Thomas C. Wilson, "Zen and the Art of CD-ROM Network License Negotiation," The Public-Access Computer Systems Review 1, no. 2: 4-14 (1990). To retrieve this article from the University of Houston's list server, send the e-mail message GET WILSON PRV1N2 F=MAIL to LISTSERV@UHUPVM1 (BITNET) or LISTSERV@UHUPVM1.UH.EDU (Internet).
5. Thomas C. Wilson and Charles W. Bailey, Jr., "The Intelligent Reference Information System CD-ROM Network," in Library LANs: Case Studies In Practice and Application, ed. Marshall Breeding (Westport, Conn.: Meckler, 1992), p. 157-71.
6. Kathleen Gunning, "The Intelligent Reference Information System: The Effect on Public Services of Implementing a CD-ROM LAN and Expert System," Library Administration & Management 6: 146-53 (Summer 1992).
7. The Quick Reference Cards developed by the IRIS Project are available from: LOEX Clearinghouse, University Library, Eastern Michigan University, Ypsilanti, MI 48197.
8. The reader unfamiliar with expert system technology may want to consult the following introductory works for further Information: Ralph Alberico and Mary Micco, Expert Systems for Reference and Information Retrieval (Westport, Conn.: Meckler, 1990); Rao Aluri and Donald E. Riggs, eds. Expert Systems In Libraries (Norwood, N.J.: Ablex, 1990); Louis E. Frenzel, Jr., Crash Course In Artificial Intelligence and Expert Systems (Indianapolis: Sams, 1987); F. W. Lancaster and Linda C. Smith, eds., Artificial Intelligence and Expert Systems: Will They Change the Library? papers presented at the 1990 Clinic on Library Applications of Data Processing, March 25-27,1990 (Urbana-Champaign, Ill.: Graduate School of Library and Information Science, University of Illinois at Urbana-Champaign); George F. Luger and William A. Stubblefield, Artificial Intelligence and the Design of Expert Systems (Redwood City, Calif.: Benjamin/Cummings Pubs., 1989); A. Morris, "Expert Systems for Library and Information Services--Review," Information Processing and Management 27, no. 6: 713-24 (1991); and Ernst J. Schuegraf, "A Survey of Expert Systems in Library and Information Science," The Canadian Journal of Information Science 15: 42-57 (Sept. 1990).
9. A detailed article about Reference Expert by Judy E. Myers, Charles W. Bailey, Jr., Jeff Fadell, Jill Hackenberg, and Thomas C. Wilson is in preparation. The working title is "Reference Expert: An Expert System for Print and CD-ROM Reference Sources."
10. James Parrott's work helped stimulate the thinking of the Knowledge Engineering Group about the structure of the knowledge base. See James R. Parrott, "Simulation of the Reference Process, Part II: REFSIM, an Implementation with Expert System and ICAI Modes," The Reference Librarian, no. 23: 153-76 (1989).
11. For further information about knowledge representation using frames, see Kamran Parsaye and Mark Chignell, Expert Systems for Experts (New York: Wiley. 1988), p. 161-210; and John R. Walters and Norman R. Nielsen, Crafting Knowledge-Based Systems: Expert Systems Made Easy/Realistic (New York: Wiley, 1988), p. 209-52.
12. Expert Systems in ARL Libraries, SPEC Kit 174 (Washington, D.C.: Office of Management Services, Association of Research Libraries, 1991), p. 2 of the flyer. Also available in ERIC (ED 337 178).
13. Charles W. Bailey, Jr., "Intelligent Library Systems: Artificial Intelligence Technology and Library Automation Systems." Advances in Library Automation and Networking 4: 11 (1991).
14. To obtain a copy of the expert system. send a stamped, self-addressed diskette mailer and a formatted 5 1/4-inch 1.2 MB or 3 1/2-inch 1.44 MB diskette to: Charles W. Bailey, Jr., Assistant Director for Systems, University Libraries, University of Houston, Houston,TX 77204-2091. If you reside outside of the U.S. and cannot get appropriate stamps, enclose $2 in your local currency to cover postage costs.
Copyright (C) 1992 by Charles W. Bailey, Jr. All Rights Reserved.