Why Are Online Catalogs Still Hard to Use?

Transcription

1 Why Are Onine Cataogs Sti Hard to Use? Christine L. Borgman Department of Library and Information Science, 2320A Moore Ha, Box , University of Caifornia, Los Angees, Los Angees, CA E-mai: We return to arguments made 10 years ago (Borgman, 1988a) that onine cataogs are difficut to use because their design does not incorporate sufficient understanding of searching behavior. The earier artice examined studies of information retrieva system searching for their impications for onine cataog design; this artice examines the impications of card cataog design for onine cataogs. With this anaysis, we hope to contribute to a better understanding of user behavior and to ay to rest thecard cataog design mode for onine cataogs. We discuss the probems with query matching systems, which were designed for skied search intermediaries rather than end-users, and the knowedge and skis they require in the informationseeking process, iustrated with exampes of searching card and onine cataogs. Searching requires conceptua knowedge of the information retrieva process-transating an information need into a searchabe query; semantic knowedge of how to impement a query in a given system-the how and when to use system features; and technica skis in executing the query-basic computing skis and the syntax of entering queries as specific search statements. In the short term, we can hep make onine cataogs easier to use through improved training and documentation that is based on information-seeking behavior, with the caveat that good training is not a substitute for good system design. Our ong term goa shoud be to design intuitive systems that require a minimum of instruction. Given the compexity of the information retrieva probem and the imited capabiities of today s systems, we are far from achieving that goa. If ibraries are to provide primary information services for the networked word, they need to put research resuts on the information-seeking process into practice in designing the next generation of onine pubic access information retrieva systems. Introduction This Specia Issue of the Journa of the American Societyfir Information Science on onine cataog research is an appropriate time to revisit the questions asked in an artice that appeared a decade ago: Why are onine cataogs hard to use? Lessons earned from informationretrieva studies (Borgman, 1986a). We caed for substantia changes in onine cataog design based on our 0 I996 John Wiey & Sons, Inc. increasing knowedge of user behavior. Despite the attention that and other artices on onine cataogs received ( Efthimiadis, 1990) itte seems to have changed. Subsequent research studies continue to report that users have great difficuty searching onine cataogs. Thus we ask the question, Why are onine cataogs sti hard to use? We argue that onine cataogs continue to be difficut to use because their design does not incorporate sufficient understanding of searching behavior. Research in information seeking indicates that users formuate questions in stages, graduay coming to the point where they can begin to articuate a query. Even then the search process may be iterative and searching may serve to refine the question rather than to buid a set of documents that matches an expicit query. A search may be conducted over a number of sessions with different information technoogies and sources, both onine and offine, picking and choosing from mutipe options to answer a question or expore an issue. Yet the design of most operationa onine cataogs assumes that users formuate a query that represents a fixed goa for the search and that each search session is independent. The first generation of onine cataogs foowed either of two query-oriented design modes: Onine card cataog modes, emuating the famiiar card cataog, or Booean searching modes, emuating information retrieva systems such as DIALOG or Medine. Secondgeneration onine cataogs merged these two design modes and improved access points, search capabiities, and dispay options ( Hidreth, 1987, 1993). Most onine cataogs currenty in use provide second-generation functionaity. The record structure, content, and primary searchabe fieds are drawn from card cataog design modes, whie the searching functions and many of the interface design characteristics are drawn from retrieva system modes. Whie user input is simper and screen dispays are much cearer and more attractive, the basic functionaity of onine cataogs has changed itte since the ate 1980s. We imit the scope of our discussion to the automated cataog JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE. 47(7): , 1996 CCC /96/

2 per se and do not address the fu range of information systems and services in which it now may be embedded. In the 1986 artice (Borgman, 1986a), we examined studies of information retrieva system searching for their impications for onine cataog design. In this artice, we examine the impications of card cataog design for onine cataogs. Athough comparing card and onine cataogs may seem ike revisiting od territory, we find this anaysis to be of current importance for severa reasons. The ibrary community is considering radica changes in cataoging codes to address the needs of networked, distributed computing environments and the description of materias in a vast array of new media. Today s onine cataogs hod many miions of records that were constructed for card cataogs and retain the same underying structure. Changes to cataoging rues and reevant standards must support the migration of od data to new formats, for ibraries have created far too much data aready to start over. Onine cataogs are an estabished technoogy in major research ibraries in deveoped Western countries, but in much of the word, card cataogs are current technoogy for a types of ibraries, and they continue to be maintained in many smaer ibraries in deveoped countries. Libraries in Centra and Eastern Europe and in many other parts of the word are making the transition directy from card cataogs to the most modern onine cataogs, having skipped the interim generations of technoogy (e.g., Borgman, 1995 ), and are faced with an array of research, training, and assistance issues in introducing these systems. Onine cataog technoogy has not kept pace with expectations for ease of use or functionaity, resuting in cas for a return to the trusted card cataog (Baker, 1994). We caim that many of the probems that remain in onine cataogs are due to the remnants of the card cataog in its structure and a faiure to design user interfaces based on the knowedge and skis of onine cataog users. We hope that the arguments presented here wi contribute to a better understanding of user behavior and ay to rest the card cataog design mode for onine cataogs once and for a. We first discuss card and onine cataogs as query design modes, then we anayze the process of formuating and executing queries in each of these types of systems. We argue that most current onine cataogs are based on card cataog design modes, that this mode does not map we onto onine systems, and that the mode is not based on information-seeking behavior. We concude with proposas for new design modes for onine cataogs. Query Design Modes Information retrieva is a difficut probem because it requires describing information that you do not yet have. Searchers must transate their information needs into a description of the information sought, reying on their own knowedge of the probem, their understanding of the toos the system provides to assist in describing the probem, and if avaiabe, the services of skied reference ibrarians or search intermediaries-the origina inteigent agents. At one extreme, information retrieva system interfaces require searchers to specify the search competey in a singe set of statements, or a query ; such systems return a set of records that match the query as a fina resuts set. At the other extreme, information retrieva system interfaces aow searchers to enter whatever fragments of the idea are avaiabe as a starting point, providing various toos to assist in exporing the information need; a query may never be stated expicity nor a fina resuts set retrieved. Most onine cataogs are based on query design modes that aow some degree of search modification but are far from being exporatory systems. Query systems were designed for highy skied searchers, usuay ibrarians, who used them frequenty, not for novices or for end-users doing their own searching. Query matching is effective ony when the search is specific, the searcher knows precisey what he or she wants, and the request can be expressed adequatey in the anguage of the system (e.g., author, tite, subject headings, descriptors, dates). Even with graphica user interfaces, the searcher must enter terms and specify reationships that match those in the database. Many onine cataogs aow users to browse authority fies and indexes, but usuay ony within the constraints of a specified query. Few systems aow searchers to retain fragments of prior search strategies and recombine them in other ways, to pursue non-inear inks in the database, or to expore by other means. Onine cataogs must serve a popuation of information seekers that is heterogeneous in terms of age, anguage, cuture, subject knowedge, and computing expertise, most of whom wi be perpetua novices at information retrieva. To design systems that can support question answering rather than simpy query matching, we need to earn more about the search process. One way to identify the features and functions to incorporate into the next generation of systems is to examine why systems based on query design modes are hard to use. The Onine Card Cataog Design Mode Current onine cataogs continue to be criticized as being more difficut to use and ess serviceabe than card cataogs. Baker ( 1994) popuarized these criticisms in a widey-discussed artice that appeared in The New Yorker, provoking support and disdain throughout the ibrary community and we beyond it. Baker s argument is based on the use of cataogs to find known books rather than to seek information or to sove information-based probems. Indeed, he caims (p. 78) that the function of a great ibrary is to store obscure books. We take the perspective that the function of a great ibrary is to assist peope in finding the information they need, whether 494 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy 1996

3 hed in that ibrary s coection or esewhere. The cataog, whether onine or hard copy, is one of severa casses of toos for finding information, and is the primary too for finding information hed in a ibrary s coection. Onine cataogs grew out of card cataogs, automating records that were designed for a manua environment, with interfaces that were intended to bring a generation of ibrary users famiiar with card cataogs into the onine word. Card cataogs remain argey structured as they were in the 19th century (Buckand, 1992). The card cataog design mode is best expained by Cutter s objects of the cataog ( Cutter, 1904, p. 12) : 1. To enabe a person to find a book of which either (a) the author (b) the tite (c) the subject is known. 2. To show what the ibrary has (d) by a given author (e) on a given subject ( f) in a given kind of iterature (poetry, drama, fiction). 3. To assist in the choice of a work (g) as to its edition (bibiographicay) (h) as to its character (iterary or topica). Based on Cutter s objects, a cataog is arranged on the assumption that searchers arrive at the cataog knowing at east one of the three access points (author, tite, or subject). However, studies of information-seeking behavior in both manua and automated environments show that peope arrive at a cataog with incompete information for any of the access points (Borgman & Siegfried, 1992; Chen & Dhar, 1990; Tagiacozzo, Kochen, & Rosenberg, 1970; Tayor, 1984). They must use information externa to the cataog (e.g., bibiographies, ists of subject headings) to obtain sufficient data to express their search within the scope of Cutter s objects. Onine cataogs based on a card cataog mode provide more searching options and may be searched with ess compete information about what is sought. However, Cutter s objects do not adequatey represent the way peope seek information. Whie Cutter certainy caimed to have the user in mind, his perspective and others of his time was based on a rationa, positivistic approach, and not on direct study of how peope formuate questions and seek information; rather, they made assumptions about the knowedge peope brought to the information-seeking process. Because card cataogs and first- and second-generation onine cataogs utiize the same data and same principes as do their predecessors, the underying record structure and access points remain the same. Accordingy, much of the search process is the same in card and onine cataogs. Onine cataogs add a ayer of functionaity, providing more techniques for searching the same data, but aso add a ayer of compexity to the process. Conceptua Knowedge, Semantic Knowedge, and Technica Skis In our earier anaysis of onine cataog searching behavior (Borgman, 1986a), we caimed that users appy two types of knowedge to the information retrieva task (p. 388): Knowedge of the mechanica aspects of searching (syntax and semantics of entering search terms, structuring a search, and negotiating through the system), and Knowedge of the conceptua aspects (the why of searching-when to use which access point, ways to narrow and broaden search resuts, aternative search paths, distinguishing between no matches due to a search error and no matches because the item is not in the database, and so on). Within this framework, a searcher usuay can achieve some resuts from the system, once the mechanica aspects of searching are conquered. Ony when the conceptua aspects are understood can the user expoit the system fuy and effectivey. Upon refection, and after 10 more years of studying onine cataog searching behavior, we see three ayers of knowedge required for onine cataog searching: Conceptua knowedge of the information retrieva process-transating an information need into a searchabe query; Semantic knowedge of how to impement a query in a given system-the how and when to use system features; Technica skis in executing the query-basic computing skis and the syntax of entering queries as specific search statements. Our 1986 anaysis merged conceptua and semantic knowedge into one category (conceptua knowedge) and regarded technica skis as mechanica knowedge. Whie a usefu dichotomy, this approach faied to make some important distinctions between users abiity to understand the overa process of information seeking that is transferrabe across systems and to understand how to conduct a search in a given system. It was not expicit in incuding the user s knowedge of the search topic in conceptua knowedge. Studies of searching behavior tend to focus on semantic knowedge and technica skis, and system evauation often considers ony technica skis, measuring users success and errors in executing search statements. Instruction in the use of onine cataogs too often focuses on technica skis aone, faiing to set those skis in the context of semantic knowedge, much ess conceptua knowedge. We anayze the knowedge and skis required to search onine cataogs in terms of these three ayers, beginning with the most genera. At each ayer, we compare the requirements of card and onine cataogs, discussing JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy

4 the knowedge and skis that transfer from one to the other and those that do not. Conceptua knowedge. Information retrieva is a difficut task, whether conducted in a card or onine cataog. The process is not aways a singe act of formuating a query; rather, it often begins with some vaguey-fet need of wanting to know something and graduay evoves to the point where one can describe some attributes of the documents that might contain the information. Once the need can be phrased sufficienty to begin searching, the question itsef may change through mutipe iterations of finding and using information resources. Thus peope usuay approach an information retrieva system with a partiay-formed query to be negotiated. The most compex part of the process often occurs offine, thinking through the probem to reach a point where it can be articuated. When searching for information, a person is seeking knowedge or meaning (e.g., what? why? how?), but must formuate a query in terms of the content (e.g., words, numbers, symbos) of information entities (e.g., documents, objects). Information retrieva systems can dea directy with information ony as things (Buckand, 1991 ), containers ( Barow, 1994 ), or entities that carry some information content that may impart meaning to its creators and users. In onine cataogs, those entities are bibiographic records. The success of a query is a function of the abiity to transate the intended meaning into a set of search terms that are contained in the bibiographic records in the cataog and that convey the intended meaning. We have yet to design methods of organizing information (e.g., cataoging, cassifying, indexing) that adequatey bridge the gap between the way a question is asked and ways it might be answered, despite the appication of sophisticated techniques to bring together terms with simiar meanings and to distinguish among mutipe meanings of a given term. Formuating author and tite queries is difficut because the searcher must have accurate information that matches the entry in the cataog or know how to identify aternative forms that might exist. Subject searching is even more difficut because the searcher must find ways to articuate his or her intended meaning using terms that match those in the cataog, whether assigned by an author, indexer, or subject cataoger. The difficuties in query formuation, particuary with regard to subject, are we-studied, and a fu anaysis is beyond the scope of this artice; we refer the reader to Bates ( 1986, 1989); Crawford, Thorn, and Powes ( 1993); Efthimiadis ( 1992, 1993); Hidreth ( 1993); Lancaster, Conne, Bishop, and McGowan ( 1991); Larson ( 1991 a, b,c); Markey ( 1986); Matthews, Lawrence, and Ferguson, 1983; O Brien (1994); Tayor (1984); Waker ( 1988); and Waker and Hancock-Beauieu ( 1991). Searchers aso need the conceptua knowedge of how search terms can be combined if they are to construct search statements with mutipe concepts. Knowedge of Booean ogic is specific to automated systems and is not transferrabe from card cataog usage. It is a difficut concept to grasp, as we have earned from studies of onine cataogs and other onine retrieva systems (Borgman, 1986a,b). Even scientists and engineers who have expertise in ogic for other appications often use and and or backward in the searching process (Borgman, Case, & Meadow, 1989). Research by cognitive psychoogists (Tversky & Kahneman, 1974) on heuristics and judgment finds that peope do not appy normative ogica modes such as Booean ogic in their everyday reasoning. Rather, they foow intuitive judgement, which incudes appying and and or in their inguistic sense-and is incusive, making things bigger; or is excusive, as in either/or, making things smaer-the reverse of the way these words act as Booean operators. Semantic knowedge. Given some conceptua knowedge of the search process, the information seeker needs the semantic knowedge of how to impement a query in a given system. Computer scientists distinguish between semantic knowedge, i.e., the meaning of computing and task concepts, and syntactic knowedge about a system, i.e., the way in which commands or actions are arranged (Shneiderman, 1992 ). We regard technica skis as syntactic knowedge in this framework. Using card cataogs requires the semantic knowedge of how a particuar cataog is structured, such as whether it has mutipe fiing sequences by time period or access point. The structure of onine cataogs is ess physicay apparent than in card cataogs. Rather than opening a card cataog drawer, onine cataog searchers must enter a search statement. In most onine cataogs, a search statement consists of three parameters in sequence: Action (e.g., find, seect, scan, browse), an access point or fied tag, (e.g., author, tite, subject), and search terms, either aone (e.g., Shakespeare, bioogy ) or in combination with Booean operators (e.g., computers and behavior, heat or therma ). Action. The action in searching a card cataog is to scan or fip through cards in sequence, after determining the access point and the set of drawers in which the appropriate cards are fied. The action in onine cataogs is the first parameter to specify, athough not necessariy the first that the searcher needs to determine. Onine cataogs typicay offer at east two actions: A find or keyword search that matches individua words or word combinations in the specified fied and retrieves a set of records; and a browse search that matches words in either the name or subject authority fie and retrieves a set of headings with a set of records inked to each heading. The distinction in function between these two actions can be difficut to expain to the perpetua novice searcher who has imited knowedge of bibiographic organization. O Brien ( 1994, p. 224) notes that whie i- 496 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy 1996

5 brarians see a vita difference between keyword and subject searching, they are as yet unsure of the roe of each. Access points. Searchers must determine which access point to use based on their assessment of the information they have avaiabe. They need to decide what fragments of each access point are most compete, distinctive, or reiabe. Card cataogs in the United States, foowing Cutter s objects, usuay offer author, tite, and subject access. The fact of three access points is more apparent in divided cataogs than in dictionary cataogs, in which the two or three sequences are interfied. Some card cataogs compicate the matter further by combining two sequences (typicay author and tite) into one cataog and keeping the third sequence in another. In Europe and esewhere, card cataogs may be searchabe by main entry ony (usuay author) and perhaps by cassification. Providing comprehensive subject access (e.g., Library of Congress Subject Headings) to cataog records is argey a U.S. practice, athough other subject thesauri or systems of organization may be appied. Most onine cataogs force the searcher to choose among author, tite, and subject access points for the initia search statement. Other fieds in the records, such as cassification numbers, anguage, series, date or date range, or type of materia, can be used to imit a search statement, but usuay must be entered either as an additiona parameter or as a subsequent command. The forms of subject access to card and onine cataogs vary widey due to oca, nationa, and regiona variations in cataoging practice, as we as to differences in onine cataog software capabiities. Search terms. The search terms may be the first parameter determined, athough not the first to be specified in card or onine cataog searching. Card cataogs and first-generation onine cataogs offer imited choices of search terms, typicay as foows: Author cataog: Surname of persona author or the first significant word in a corporate name; Tite cataog: First significant word in the tite; Subject cataog: First significant word in subject headings or index terms; Shef ist: Cassification number or cass mark. Cassification access to card cataogs rarey is offered to the pubic in the U.S. Shef ist cataogs are essentia for cataog maintenance but usuay are kept in staff areas. Cassification access is more common in Europe and may be offered in pace of the subject cataogs that are typica in the U.S. Search term options increased greaty with the introduction of second-generation onine cataogs, which aow keyword searching of the access points. Current systems usuay offer the foowing options: Author search: Surname and given name of persona author or any significant word in a corporate name; Tite search: Any significant word in the tite; Subject search: Any significant word in subject headings or index terms; Shef ist: Cassification number or cass mark, or a range of numbers. Searchers need to know whether they can enter the cataog ony under the search terms corresponding to the first significant word in the access point or whether a words are searchabe. If ony the first word is searchabe, searchers must be abe to determine which word that might be. Variations in cataoging practice resut in inconsistent entries, both among different cataogs and within any given cataog. For exampe, the author cataog may incude surnames of a authors or ony first authors of a work; determining the appropriate search terms for corporate authors is even more difficut. Searchers may need to know a tite precisey enough to know the first significant word and understand the notion of stop words, i.e., words deemed insignificant or too common to be usefu for fiing, typicay artices. Identifying search terms for the subject cataog is hardest of a, since peope often do not recognize that the subject entries are drawn from a controed ist or thesaurus that is separatey searchabe itsef. Instead, they enter the cataog using the free-text keywords they know best, often on a tria-and-error basis. Keyword searching of subject headings provides a powerfu mechanism to ocate a headings containing a given word and a records containing a given word in the subject fieds. However, it does so at the expense of the structure of the subject heading system, which was designed to provide context with headings and subheadings. Many researchers have addressed the probem of subject access searching and dispay mechanisms in onine cataogs, incuding Bates ( 1989), Borgman, Gaagher, Hirsh and Water ( 1995 ), Leazer ( 1994)) Markey and Demeyer ( 1986), McGarry and Svenonius ( 1991 ), Micco ( 1991 ), Rosenberg and Borgman (1992),andTiett(1991). Booean ogic. In card cataogs, terms must be searched individuay. In onine cataogs, search terms may be combined with Booean operators, whether the operators are stated expicity by the searcher or are impied by the system. Specifying queries that contain Booean ogic requires conceptua knowedge of how search terms can be combined and semantic knowedge of how a given system executes Booean operators. Even if a searcher understands the concepts of combining terms with operators, anomaies in system features may interfere with empoying that knowedge. Onine cataogs vary widey in how they treat operators, particuary in the use of impicit operators and in the order they execute them. Impicit Booean and execution order are JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy

6 difficut features to expain and few systems document them ceary. We provide severa exampes to iustrate the compexity of these concepts and the significance of their effect on search resuts. Eary onine cataogs aowed singe-word search terms ony; mutipe-word search terms required expicit Booean operators to specify how the words shoud be treated. Second-generation systems typicay aow the searcher to enter mutipe-word phrases, which simpifies the input process but eaves to the system the decision on how to combine the words. Some systems appy an impicit OR between words, others appy an impicit AND, and yet others treat mutipe-word search terms as phrases bound together in that sequence. Thus, a search for research design woud yied a arge set of a records containing either of these genera words, on a system that empoys impicit OR; a sma set of records containing both of these words, in a system that empoys impicit AND; and a sti smaer set of records in a system that treats the words as a phrase in the specified sequence. Many commercia onine retrieva systems execute Booean operators in agebraic sequence, e.g., reading the whoe string and first executing statements within parentheses, then NOT, AND, and OR. Most onine cataogs do not aow parentheses and simpy execute words and operators from eft to right, ignoring the agebraic hierarchy of operators. Thus the same phrase wi receive very different resuts depending on the sequence of operators and the order in which they are executed. To iustrate this difference, et us compare the queries behavior or psychoogy and computers and computers and behavior or psychoogy. We assume the searcher intends the same itera meaning for these phrases, which is to combine computers with any records containing either behavior or psychoogy. Most onine cataogs woud treat the first query as the user intended [(behavior OR psychoogy) AND computers], whie any system that executes operators in agebraic order woud not, instead retrieving [behavior OR (psychoogy AND computers)]. The second phrasing woud retrieve unintended resuts whether the system executes eft to right or in agebraic order [(computers AND behavior) OR psychoogy] ; it woud retrieve the intended resuts ony in those systems that read the fu phrase and then execute OR before AND. Ony the most sophisticated searchers are sufficienty knowedgeabe of Booean ogic to seek documentation on how the system treats muti-word search terms and the order in which it executes operators. If they devote enough time and effort, they can experiment with mutipe search statements that wi revea the system features. However, even expert searchers may achieve confusing, incompete, or misrepresentative resuts by assuming that the system operates in a fashion simiar to other systems with which they are famiiar. Peope carry over knowedge from one automated system to the next, just as they carry knowedge of card cataogs to onine cataogs (Borgman, 1986b; Gentner & Stevens, 1983). Fie organization. Once some resuts are retrieved, the searcher must ocate the records in the fiing sequence, which requires some semantic knowedge of fie organization, Without further instruction, peope assume cataogs are arranged ike a teephone book or some other we-known fiing sequence. However, card cataogs typicay are fied by an eaborate set of rues known ony to cataogers and para-professiona cataog fiers-various editions of the American Library Association Fiing Rues range from 50 to 260 pages in ength! Some of the features of the fiing rues can be expained in simpe terms, such as the difference between word-byword fiing and etter-by-etter fiing, but the rea difficuties ie in the fact that cards may be fied based on information that is not on the card, such as whether the entry is treating a person as an author or a subject, or the fu speing of a word abbreviated on the card (e.g., cards for Engish works starting with St. (for Saint) may fie before cards for German works starting with St. (for Sankt)). Because the fiing rues are so compex, cards often are misfied, eading to custers of cards correcty fied behind one misfied card. Thus, even if the correct access points and search terms are known, it may be difficut to ocate the cards, or to determine if the ibrary indeed has any books that match the searched terms. Just as cards are fied in a specified aphanumeric sequence, onine cataog records are dispayed in a specified aphanumeric sequence. However, the sorting sequence is programmed based on data in the record. This means that the ALA Fiing Rues cannot be fuy repicated because they rey on information externa to the record that is suppied by the human fier (athough ibrarians went to great effort to repicate these fiing rues, in the eary years of onine cataogs, before succumbing to a machine fiing sequence). As a resut, most onine cataogs sort cataog records in a sequence that is more simiar to a phone book than a card cataog. Onine cataogs based on MARC records have made some improvements over straight aphanumeric sorting, such as coding access points to identify the first significant word for fiing, which aows fieds beginning with a stop word (e.g., The Story of 0 ) to fie under the first significant word ( story ) but aows records to fie under stop words when necessary (e.g., A is for Appe shoud fie under A, not is or Appe ). In addition, most of the commerciay-produced onine cataogs presenty in use were deveoped in the Engish-speaking word and incorporate ony a few characters beyond the standard The first edition of the ALA Fiing Rzdes (American Library Association, 1942) is 109 pages; the second edition (American Library Association. 1968a) is 260 pages; the second edition abridged (American Library Association, 1968b) is 94 pages; and the 1980 edition (American Library Association, 1980) that is the successor to the prior editions was reduced to 50 pages. The compexity of these rues aso is apparent from the existence of substantia training manuas, such as Carothers ( 198 1) at 120 pages and Hoffman ( 1976) at 176 pages. 496 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy 1996

7 Engish aphabet. The rest of the characters are ost in transiteration or in omitting diacritics. As the non-engish-speaking word comes onine and preserves their fu character sets in their onine cataogs and other retrieva systems, machine fiing order, keyboard input, and dispay, wi become ever more compex (e.g., Peruginei, Bergamin, & Ammendoa, 1992). Machine fiing order probaby is more intuitive than are the ALA Fiing Rues because it reies on data in the record and because the sequence is easier to see when 10 to 20 items are dispayed on a screen at once. Stricty speaking, records never are fied out of sequence in an onine cataog, eiminating one of the probems of the card cataog. However, a number of other factors can make it difficut to scan a ist of records for the item sought. For exampe, searchers have difficuty with the sort order of mixed etter and number sequences in ca numbers, both on the screen and on the shef. Which comes first, or Z699.6? Numericay, 6 is much ess than 51; but aphabeticay, 5 sorts before 6. In addition, typos (e.g., Untied States ), variant punctuation (e.g., U.S. vs. US), and other variations in entry form (e.g., addresses, essays, and ectures, addresses, essays, & ectures, AEL ) introduce a the fiing probems of card cataogs and a few new ones. Probems with sort order and with stop words presumaby are ess in onine cataogs than in card cataogs because more aternatives exist for ocating items. Variant forms, once identified, are more easiy corrected in automated systems than in manua systems due to search and repace functions-in some systems one action can change a occurrences of Untied States to United States, for exampe, where such a goba change woud be prohibitivey expensive in a card cataog. Technica skis. Given the semantic knowedge of how to impement a query in a given system, the searcher needs the technica skis to formuate and execute specific search statements. The primary technica ski required by card cataogs is the abiity to manipuate cards in the drawers. The organization of the card cataog is physicay apparent, particuary in a divided cataog, and a searcher seects a starting point by puing out a drawer in the aphabetic vicinity of the desired known item or subject heading. Part of the appea of card cataogs, that which Baker ( 1994) captures so we, is the tactie sense of fipping through cards, wandering through the ibrary s coection. Onine cataogs require technica skis that card cataogs do not, such as basic computing concepts and the syntax of commands or actions in the system. Using computer terminas or work stations. We must not take for granted that a ibrary users are computer users, or that their knowedge of computers is adequate or appropriate for onine cataog use. A recent artice in the Wa Street Journa (Carton, 1994) describes some of the naive questions directed to the hep ines operated by computer hardware and software companies. The anecdotes are instructive for ibrarians: A customer trying to operate a mouse as a foot peda, others pointing the mouse at the screen rather than roing it on a pad, some who have difficuty finding the power switch, and many customers who request the ocation of the any key in response to screen instructions to press any key. And these are the peope who do have a computer! Whie searchers of onine cataogs need not be computer experts, they must be famiiar with the ayout of the keyboard, both the keys in common with typewriters (etters, numbers, space bar, caps ock, etc.) and keys specific to computers (contro, return/enter, deete, function keys, etc.), and with conventions such as pressing return/enter after typing a command and using arrow keys or a mouse to move about the screen. Often searchers must earn system-specific function keys, or specia uses of genera keys, such as tab for a new ine, or differences between enter and return on systems that use these keys for different purposes. Searchers aso must understand screen dispay conventions such as consistent pacement of instructions, messages to the user, and user input, as we as the nature of menu seection. These basic skis rarey are incuded in the instruction manuas for onine cataogs, much ess the reference sheet. For the skied computer user, this information may seem trivia. For the searcher with no prior computer experience, the ack of these skis may prove to be an insurmountabe barrier since it is difficut to know where to start. Onine cataogs were introduced in ibraries ong before computers were widey impemented in schoos, work paces, and homes. Whie severa experimenta onine cataogs existed in the 1960s and eary 1970s it is generay acknowedged that the first arge-scae impementations were at Ohio State University in 1975 (Mier, 1979) and the Daas Pubic Library in 1978 ( Borgman, 1978; Borgman & Kaske, 1980 ). By the eary 1980s a sufficient number of onine cataogs were in pace in the United States for the Counci on Library Resources to commission a major study of onine cataog usage (Matthews et a., 1983). Time-sharing systems appeared in the 1960s and hobbyist persona computers in the ate 1970s (Press, 1993)) but arge-scae desktop computing is a 1980s phenomenon-the IBM persona computer was introduced in 1981 and the Appe Macintosh in Ony in the ast few years has it become reasonabe to assume that many ibrary users in the West are famiiar with computers. In other parts of the word, where onine cataogs are being introduced in societies with far ess information technoogy in pace, we cannot make this assumption. For many peope, ibraries wi continue to provide their first encounter with computers. Syntax of entering search commands. Onine cataogs provide structure in more abstract ways than do card cataogs, and the starting point for browsing must be JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy

8 stated much more expicity than puing out a random drawer. Most onine cataogs require the user to enter commands in a precise format and sequence such as the three-parameter structure discussed above to describe the semantic knowedge requirements. This structure is fairy consistent across command-driven onine cataogs, athough the terminoogy varies widey among systems (uness the Common Command Language is impemented, but it has not been widey adopted in the U.S.), In these exampes we use FIND for action, often abbreviated as FI, and AUTHOR (AU) for author. The abe for the author fied varies widey and can be found aso as AUT, NA (name), or PA (persona author) among others. The format for entering the search terms varies, with some systems being more rigid than others. For exampe, to search for works by Robert M. Hayes, one needs to know which, if any, of the foowing search statements are acceptabe to the system. Some may match a works in the cataog by this author, some a subset, and others none at a. FI AU ROBERT M. HAYES FI AU R M HAYES FI AU HAYES, ROBERT M FI AU HAYES, ROBERT M. FI AU HAYES ROBERT M FI AU HAYES ROBERT FI AU HAYES, ROBERT FI AU HAYES RM FI AU HAYES R.M. FI AU HAYES R FI AU HAYES, R? The search statements in this exampe vary by word order, number of words in the search terms, competeness of search term (name or initias), and punctuation; none incude expicit Booean operators. Most newer onine cataogs are fairy forgiving on these factors, searching words in author names as Booean combinations in any order and ignoring punctuation, whie oder systems (but sti in genera use) require precise character-bycharacter matches. The ast entry in the above ist is an exampe of truncation, which aows for variant word endings. This feature is simiar to fipping through cataog cards before and after the one sought to ook for variant endings. The distinction between exact-match and browsing searches in onine cataogs can be difficut to expain. Most onine cataogs match exacty on the characters entered and do not seect variant endings uness expicity tod to do so. The person with minima computer experience may not reaize that an author search for HAYES R wi not match HAYES ROBERT. To retrieve a variant endings may require a different command that searches the authority fies, such as a browse search discussed previousy. Commands to manipuate thesystem. In addition to the commands required to execute a query, onine cataogs have commands to contro various aspects of the system. We divide these into severa groups: Commands to contro the search. Onine cataogs incude a variety of contro commands, such as actions to modify a prior search, start a new search, move forward and backward through ists, and switch between ists of subject headings and ists of bibiographic records. Commands to contro dispay formats. Card cataogs have ony one dispay format: The card. Onine cataogs often have severa types of dispays under searcher contro. Search resuts may be presented as a summary ist of singe-ine entries, and individua records may be dispayed as short or ong versions of bibiographic records that vary by which fieds are shown (e.g., notes, hodings, circuation status). Commands to contro output of resuts. One of the weaknesses of eary onine cataogs was that resuts sti had to be copied by hand onto sips of paper, offering itte improvement over card cataogs. Now many onine cataogs aow search resuts to be printed. downoaded onto disks, or even sent by eectronic mai. Various commands aow the user to contro how this output is formatted and where it is sent. Making Onine Cataogs Easier to Use Onine cataogs shoud be judged by their success in answering questions rather than by their success in matching queries. In the ong term, we need to design systems that are based on behaviora modes of how peope ask questions. Such a design mode coud assist in the question-negotiating process, aowing the searcher to pursue mutipe avenues of inquiry by entering fragments of the question, exporing vocabuary structures, capturing partia resuts, reformuating the search with the assistance of various speciaized inteigent agents, retaining eements of a search for future sessions, and even transferring eements to other systems. Many researchers are proposing new modes for the design of onine cataogs, such as Bates ( 1989) berry-picking and (Bates, 1986 ) subject access design mode : Hidreth s ( 1993 ) mode of the E 30PAC : Enhanced, expanded, and extended onine pubic access cataog; and the work of Robertson, Hancock-Beauieu, and Waker (Robertson & Hancock-Beauieu, 1992; Hancock-Beauieu, 1989; 199 1; Waker & Hancock-Beauieu, 199 1) and Efthimiadis ( 1992, 1993, 1995 ) on query reformuation and reevance ranking. We have devoted more than 5 years to the iterative design of an exporatory onine cataog based on chidren s deveopment eves, knowedge, and interests (Borgman et a., 1995). Other researchers are studying searching behavior on extant systems that provide a basis for design modes, such as the work of Bates ( 1990); Bekin, Oddy, and Brooks ( 1982); Eis ( 1989); Ingwersen (1984); Kuhthau (1988, 1991); Lynch 500 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy1996

9 ( 1992); Marchionini ( 1992, 1995); Markey ( 1980, 1986, 1989); and Shute and Smith ( 1993). Very itte of this body of research has informed the design modes of the commercia onine cataogs that are in genera use around the word. In the short term, we can hep make onine cataogs easier to use through improved training and documentation that is based on information-seeking behavior. Current instructiona approaches tend to focus on the procedures for query formuation, rather than on the question being asked. Instruction is short and precise, with exampes of how to enter an author, tite, or subject query, and exampes of Booean operators. The formats usuay consist of reference cards, onine hep, and short training sessions for the few wiing to attend. Fu instructiona manuas aimed at the end user are rare, thus documentation on important features such as impicit Booean and execution order is hard to ocate. Even harder to find is a conceptua framework for the how, when, and why of searching. Some peope need and want more instruction than do others, and no singe form of instruction wi be adequate or appropriate for a onine cataog users. Onine cataogs wi be easier for most peope to use if they can find a eve and type of instruction or documentation that suits their needs. Some wi prefer sets of documentation simiar to that provided for other compex software such as word processing or spreadsheet appications, incuding instructiona manuas, reference manuas, and summary reference cards. Others wi prefer short summaries that provide an overview and a set of exampes. Sti others, incuding skied searchers famiiar with many retrieva systems, wi want detaied documentation on how the system handes each feature and function. Given the increasing portion of users who access onine cataogs remotey rather than from the ibrary buiding, much of the instructiona materia needs to be onine as part of the system. Those who use terminas in the buiding may be served best by aides who can provide individua assistance and who, in turn, can troubeshoot system probems. Searching assistance can be provided to remote users by an onine heper who can respond to questions in rea time. High-quaity instruction can overcome some probems in system design, but shoud not be a substitute for design improvements. Time invested in eaborate hep systems often is better spent in redesigning the user interface so that the hep is no onger needed. We earned eary in our studies of onine cataogs that peope were not wiing to devote much time to earning to use these systems (Matthews et a., 1983). Searchers expect systems to be easy to use with reativey itte time investment, which has considerabe impications for ibrary services. As O Brien ( 1994) notes, we assumed that serious users of card cataogs woud invest considerabe effort in earning to use them we; we have fewer expectations of onine cataog users. Thus, we are not caiming that card cataogs were easier to use or required ess instruction. Rather, we have higher expectations of onine cataogs capabiity to support the information-seeking process than we did of card cataogs. We aso have devoted more study to understanding the onine cataog searching process, so we have better data on the probems. Our ong-term goa shoud be to design intuitive systems that require a minimum of instruction. Given the compexity of the information retrieva probem and the imited capabiities of today s systems, we are far from achieving that goa. Summary and Concusions Despite numerous improvements to the user interface of onine cataogs in recent years, searchers sti find them hard to use. Most of the improvements are in surface features rather than in the core functionaity. We see itte evidence that our research on searching behavior studies has infuenced onine cataog design. Whie queries now may be simper to input, searchers sti bear the burden of transating their question into a precise structure that the system can interpret. Onine cataogs continue to require searchers to specify a query in terms of actions, access points, search terms, and Booean operators to begin a search. Query-based systems were designed for expert ibrarian searchers who have a rich conceptua framework for information retrieva; their expertise ies in transating questions into queries on behaf of end users. Most end users of onine cataogs are perpetua novices who ack the requisite conceptua knowedge for searching. They need assistance in the transation process, whether provided by the system itsef, by instruction in using the system, or by a search intermediary. Further improvements in the user interfaces to querybased retrieva systems are ikey to have minima effect on searchers abiities to answer questions. Instead, we need to incorporate more knowedge of searching behavior into the design of these systems. Onine cataogs shoud assist searchers in empoying whatever information they have in hand to obtain an answer to their question. System designers must recognize that the goa of a search may be merey to refine the question, not to obtain an answer, and that a search may be conducted over mutipe sessions and mutipe systems. Assistance and expanation shoud be offered accordingy. Current systems can be made more effective through training that provides a conceptua framework for searching rather than simpy a set of procedures for stating queries, with the caveat that good training is not a substitute for good system design. Onine cataogs are the most widey-avaiabe automated retrieva systems and the first that many peope encounter. They have become the core of pubic access systems that provide a range of databases and other information services. After neary 30 years of experience JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy

10 with commercia retrieva systems and 20 years with onine cataogs, we woud expect them to empoy the atest in interface design, yet we continue to find them difficut to use. As software for other educationa, business, and entertainment appications becomes simper and more powerfu, expectations of the onine-cataog-user community wi continue to increase, pacing even greater demands on the ibrary community to provide systems that are at east as refined as today s desktop workstations. If ibraries are to provide primary information services for the networked word, they need to rise to this chaenge, putting research resuts on the information-seeking process into practice in designing the next generation of onine pubic access information retrieva systems. Acknowedgments Portions of this artice are based on a tak presented in the Wayne State University Libraries Distinguished Lecture Series in Apri The author is most gratefu for extensive commentary on earier drafts of the paper by Donad Case, Davina Kein, and Katain Radics of UCLA, and Chares Hidreth of the University of Okahoma. Nadia Caidi of UCLA provided expert bibiographic assistance. A errors, omissions, and interpretations remain the responsibiity of the author, of course. References American Library Association. ( 1942). A.L.A. ru~js./i~r,fiing cataog card.x Prepared by a specia committee; Sophie K. Hiss, chairman. Chicago: American Library Association. American Library Association. Fiing Committee. ( 1980). ALAjiing rues. Chicago: American Library Association. American Library Association. Subcommittee on the ALA Rues for Fiing Cataog Cards. ( 1968a). ALA ru/es,for,fi/ing cafuog curds. Prepared by the ALA Editoria Committee s Subcommittee on the ALA Rues for Fiing Cataog Cards. Pauine A. Seey, chairman and editor (2nd ed.). Chicago: American Library Association. American Library Association. Subcommittee on the ALA Rues for Fiing Cataog Cards. ( 1968b). AL4 rues for jing cafaog cards. Prepared by the ALA Editoria Committee s Subcommittee on the ALA Rues for Fiing Cataog Cards. Pauine A. Seey, chairman and editor (2nd ed.. abridged). Chicago: American Library Association. Baker, N. ( 1994, Apri 4). Discards. Theh ew Yorker passim. Barow, J. P. ( 1994). The new economy of ideas. Wired, Bates. M. J. ( 1986). Subject access in onine cataogs: A design mode. Journa of the American Society jijr I&m&on Science, 37, Bates. M. J. ( 1989). The design of browsing and berry-picking techniques for the onine search interface. Onine Review. Z3( 5), Bates, M. J. ( 1990). Where shoud the person stop and the information search interface start? It$ormatiun Processing & Management, 26(5), Bekin, N. J., Oddy, R. N.. 81 Brooks, H. M. ( 1982). ASK for information retrieva: Part 1. Background and theory. Journa qfdocumencation, 38(2), Borgman, C. L. ( 1978). The roe of technoogy.for the Daas Pubic Library in ong range panning. Educationa Resources Information Center, September (ERIC Document Reproduction Service No. ED I53 698) Borgman, C. L. ( 1986a). Why are onine cataogs hard to use? Lessons earned from information retrieva studies. Journa qfthe American Society.fi,r Ir$wmation Science, 37( 6) Borgman, C. L. ( ). The user s menta mode of an information retrieva system: An experiment on a prototype onine cataog. Internationa Journa (2/ Man-Machine Studies, 24( 1) Borgman, C. L. ( I995 ), Information retrieva or information morass? Impications of ibrary automation and computing networks in centra and eastern Europe for the creation of a goba information infrastructure. In T. Kinney (Ed.), Proceedings the Annua Meeting of the 4merican Society.for Ifformation Science, 32, October 9-12, 1995, Chicago (pp ). Medford, NJ: Information Today. Borgman, C. L., Case, D. 0.. & Meadow, C. T. ( 1989). The design and evauation of a front end user interface for energy researchers. Journa ofthe. mericun Societyfor Information Science, 40, Borgman, C. L., Gaagher, A. L., Hirsh, A. G., & Water, V. A. ( 1995 ). Chidren s searching behavior on browsing and keyword onine cataogs: The Science Library Cataog Project. Juurna oythe,4merican Society& It@mation Science. 46, Borgman. C. L., & Kaske. N. K. ( 1980). Onine cataogs in the pubic ibrary: A study to determine the number of terminas required for pubic access. Proceedings ofthe 43rd American Societyfor Information Science Annua /Meeting. October 4-9, 1980, Anaheim, CA (pp ). White Pains, NY: Knowedge Industry Pubications. Borgman, C. L., &Siegfried, S. L. ( 1992). Getty s SynonameTM and its cousins: A survey of appications of persona1 name matching agorithms. Journa of the imerican Societyfor Irzfbrmation Science, 43, Buckand, M. K. ( 1991). Information as thing. Journa of the American Society.ftir Infixmation Science, 42, 35 I-360. Buckand. M. ( I992 ). Redesigning ibrury services: A man~tisto. Chicago: American Library Association. Carton, J. ( 1994, March I ). Befudded PC users food hep ines, and no question seems to be too basic. UhStreec Journa. Carothers, D. F. ( I98 I ). Se!Jrinstruction manuaji,r the,jiing ofcataog curds. Chicago: American Library Association. Chen, H., & Dhar, V. ( 1990). User misconceptions of information retrieva systems. Internationa Journa ofman-machine Studies, 32, Crawford, J. C., Thorn, L. C., & Powes, J. A. ( 1993). A survey of subject access to academic ibrary cataogues in Great Britain. Journa o~librarinnship and Infurmation Science, 25( 2), Cutter, C. A. ( 1904). Rues& a dictionary cutaog(4th ed). Washington, DC: Government Printing Office. Efthimiadis, E. N. ( 1990). Onine pubic access cataogs: Characteristics of the iterature. Journa of Injtirmation Science, Principes. and Practice, 16(2) Efthimiadis, E. N. ( 1992). Interactive query e.xpansion and reevance jkdback./i)r documen retrieva systems. Unpubished doctora dissertation, City University, London. U.K. Efthimiadis, E. N. ( 1993). A user-centered evauation of ranking agorithms for interactive query expansion. In R. Korfhage, E. Rasmussen, & P. Wiett (Eds.), 16th Internationa ConjL;rence of the Association qf Computing Machinery, Specia Interest Group on Information Retrieva, June 1993, Pittsburgh, PA, (pp ). New York: ACM Press. Efthimiadis. E. N. ( 1995). User choices: A new yardstick for the evauation of ranking agorithms for interactive query expansion. Irzformation Processing and Management, 31(4), Eis, D. ( 1989). A behavioura mode for information retrieva system design. Journa of Injkmation Science, 15, Gentner, D., & Stevens, A. L. (Eds.). ( 1983). Mentamodes. Hisdae. NJ: Erbaum. Hancock-Beauieu, M. M. ( 1989). Onine cataogues: A case for the user. In The onine cataogue: Deveopments and directions (pp ). London: Library Association. 502 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-Juy 1996