prefer to use natural-language queries.
Libraries also handle phrase searching in different ways. Phrase searching usually is embedded within keyword search either explicitly or implicitly depending upon the ILS system. Aleph (Ex Libris) libraries use a radio button for “word or phrase” or “words adjacent” or “exact phrase” options for the computer to execute the command. Unicorn (Sirsi/Dynix) libraries provide three options: “keyword,” “begins with,” and “exact.” Some libraries have the “exact” command executed to search every field in a bibliographic record; other librar-ies search the title, subject, and author fields only. The Millennium system’s (Innovative) keyword search fea-ture can do automatic phrase and “AND” search. Some Millennium libraries (e.g., Michigan State University) take advantage of this feature to search words entered as phrases first and, if unsuccessful, the system then repeats the search for the same words using the Boolean opera-tor “AND.” This feature produces more relevant search results. However, several Millennium libraries have not implemented this feature. They still use “Boolean key-word” search as the default and instruct users to add quotation marks to define phrases. The Voyager (Ex Libris, formerly Endeavor) system offers two types of keyword searches: “keyword Relevance” and “keyword Boolean.” Both options can handle phrase searching. But users are required to enter quotation marks for specific terms used as phrases. Some libraries intentionally made only one keyword search option available. Other librar-ies provided both options and used different languages as an OPAC search key (see appendix B). These search keys are not self-explanatory, and users will often find them puzzling. The default help screen provided by the ILS vendor and adopted by many Voyager libraries does not help much either (see appendix C). Thirty-one of the 35 Voyager libraries provide a Boolean keyword search option. Only five libraries utilize the automatic “AND” feature. One library uses Boolean keyword search as the only keyword option, but did not activate the automatic “AND” functionality.
Relevance ranking in search results
When users search by keyword, the best way to sort the results is by relevance. Presenting the most relevant results at the top of the results page is crucial because it enhances library resource discovery and access. Other sorting options, such as title or publication date, are not very useful since users usually do not have titles or pub-lication dates in mind when browsing search results from a keyword search.
Three ILS systems (Millennium, Unicorn, and Voyager) have a relevance-ranking feature, yet this functionality was very much underutilized by the libraries studied. Of the eighteen Unicorn libraries, only five offered relevance
ranking. None made it the default sorting option. Thirty-six of the 38 Millennium libraries provided relevance ranking as a sorting option. Only twelve of those librar-ies made relevance ranking the default sorting system. Twenty-seven out of the thirty-five Voyager libraries offered the keyword (relevance) search option, under which the search results were automatically ranked by relevance. Out of the twenty-nine Voyager libraries that offered the keyword (Boolean) search option, only four libraries used relevance as the default sorting system. The rest of the libraries used a “system sort” mechanism that sorted search results by bibliographic control num-ber. Figure 1 summarizes the sorting options used by the ARL libraries studied and also shows the default sorting options for keyword search.
Unlike online search engines, which pull data directly from full-text documents, library OPACs search for words from the structured metadata entered by catalog-ers. Different fields are set to carry different weights for relevance considerations. The behind-the-scenes algo-rithm (the criteria used to decide the level of relevance) should be carefully established to warrant a good ranking scheme. For example, the new OPAC of North Carolina State University Library, powered by Endeca, adopted an algorithm based on field weighting, phrase matching, facet LCSH, term frequency (TF), and inverse document frequency (IDF). Their search results are indeed more logically ranked by relevance. Recently there have been suggestions to incorporate circulation statistics, book review data, and a Library of Congress call number table into the algorithm. The checkout data would provide a rough substitute for Google’s PageRank (a count of links to a site, which is an indication of the site’s popularity), and book reviews would provide more text to be consid-ered in the relevancy tests. Using Library of Congress call numbers would either require having the call number table loaded and then running the search terms against it or including call numbers in the algorithm, giving more weight to titles having the same call number. For example, seven out of twenty-three results generated for a search for “New York history” on an OPAC have the call number “F128.” The call number “F128” is linked to the call number table with the subject New York and history. It can be confirmed that seven items with call number “F128” should be considered more relevant and ranked first on the results list. More research needs to be done in this area.
the search results display
The search results display is critical. The information, options, and bibliographic data presented on the browse page help users decide what actions to take next. In the OPACs examined, the authors found the following problems:
8 inFoRmation tECHnoLoGY anD LiBRaRiEs | maRCH 2008
热门文档