A number of studies have been published investigating end user tagging of information resources. Folksonomies are the vocabularies that result from the act of users’ application of subject terms, or tags, to particular items using their own vocabulary and understanding of, or relationship to, the object.³ In contrast, LCSH and other controlled vocabularies that traditionally appear in library catalogs are based on carefully developed principles of thesaurus construction and are applied according to established standards and rules so that, ideally, terms are assigned consistently and accurately to aid the search process. Studies of inter-indexer consistency, however, have demonstrated that this ideal is rarely met.⁴ Another perceived benefit of controlled vocabularies is the disambiguation of terms, so that words with multiple meanings are understood to refer to a single definition. Unlike tagging, cataloging rules and local practices tend to limit the number of subject headings to be applied to a given item. Broader or narrower headings are applied according to the number of topics covered in a work, and subject headings are typically not assigned for topics comprising less than 20 percent of the text.

Conversely, free-text keywords in the form of tags are based on whatever, and how many, terms an end user feels are appropriate and meaningful for his or her personal use. Terminology in folksonomies may vary widely based on users’ personal vocabularies, cultural or geographic backgrounds, levels of expertise, or particular interest in the item. Terms may be applied inconsistently by different users because of misspellings, use of plural or singular forms, or capitalization. Tag terms could be much broader or narrower than the controlled vocabulary terms assigned.⁵ Many of these disadvantages could have a positive side for information retrieval purposes in some circumstances. End users may be using language that is more current than the controlled vocabulary, more specialized, or more targeted to the layperson. By providing terminology from many levels of specificity, different users could be helped by different terms.⁶ A clear disadvantage of tagging, at least in terms of search enhancement, is that since taggers primarily tag for their personal use, they might use tags that are meaningless to others.⁷ These types of tags are often referred to as personal tags; examples include nonstandard abbreviations or codes, information about the location or particular details of the user's own copy, or opinion-based terms, such as “favorite” or “boring.”

Even if some users are uninterested in providing their own tags to library catalog records, they might be helped by the presence of tags assigned by other users. If tags are indexed for searching, they can provide additional keyword access to items, using terms that may not appear in a record's description or controlled subject headings. Once a user retrieves a record, viewing a tag cloud (which presents an array of terms associated with an item) might help the user determine whether a resource meets his or her interests or needs. In addition, the presence of tags in a record might influence other users to begin tagging.⁸ To this end, libraries may choose to import user tags into their local catalogs from larger data aggregations. Rather than relying on their own users to tag items, a ready-made folksonomy can be supplied to supplement the data in the catalog record. One such source of user tags is LibraryThing for Libraries (LTFL) (www.librarything.com/forlibraries), a commercial product (now owned by Bowker) that was developed by LibraryThing for libraries to use with their existing library systems.

LibraryThing (www.librarything.com) is a social networking site that allows individual users to catalog their own book collections. Members can add tags and reviews to records for books, as well as engage in online discussions and other interactive activities. As of May 2011, 1,343,647 LibraryThing members had cataloged more than 62 million items. The personal copies cataloged by users represent 6,106,556 works, to which 76,008,376 user tags have been added, or an average of 12 tags per record.⁹ Some works generate hundreds of tags while others have none or only a few. LTFL is a separate, fee-based service offering some of the social features of LibraryThing to enhance local library catalogs, including tags, reviews, recommendations, and a browsable virtual bookshelf display of book cover images. To resolve the issue of personal tags, “preselected LTFL tags have been approved for usefulness and appropriateness by LibraryThing librarians. Highly personal tags (to read, gift from mom) have been excluded.”¹⁰ LTFL (citing March 2009 data) claimed an average 75 percent overlap with titles in the catalogs of its public library clients, but admitted that due to the nature of their collections, academic, foreign and specialized libraries would tend to have a lower overlap rate.¹¹

In a pilot study conducted in 2008, the author of the present paper selected a random sample of 383 bibliographic records from a large academic library catalog and searched the titles in LibraryThing.¹² Because many of the works searched were specialized, scholarly, foreign, or older materials (as is typical in a research library), only 21 percent of the titles were found in LibraryThing, which favors newer, more popular works in English as well as canonical literary works. However, the hit rate within the original sample for creative works of literature was 45 percent. Match percentages were even greater when the literature sample was limited by publication date or language. Fifty-four percent of literary works in English were found in LibraryThing, as was 61 percent of twentieth- or twenty-first-century literature in any language. For twentieth- and twenty-first-century literature in English, records were found for 68 percent of the sample. Because the number of literary works in the sample was relatively small (only 49 titles), further investigation is needed to determine whether this trend is representative of a larger population.

Traditionally, libraries have paid less attention to providing subject access to literary works than to nonfiction works. Part of this is practical because the “aboutness” of literary works is often more subjective than for nonfiction. Depending on the work, settings, historical periods, or characters, recurring fictional characters, and genres can be relatively simple to identify for works of fiction or drama, but thematic topics can be more elusive and more challenging to determine without reading the entire work.¹³ Even then, themes such as alienation, redemption, or betrayal may be open to interpretation and are rarely explicitly stated.¹⁴ By its very nature, poetry in most cases defies easy classification by subject, and poetry collections often lack an overarching theme to which subject terms can be applied. In 1990, the American Library Association (ALA) published Guidelines on Subject Access to Individual Works of Fiction, Drama, etc., providing a thesaurus of genre terms for literary works.¹⁵ In 1991, the OCLC/LC Fiction Project commenced in which the two sponsoring organizations collaborated with six academic and public libraries to add subject and genre headings to bibliographic records for fiction. The OCLC and LC support of the project ended in 1999, although some libraries have continued a policy of assigning subject headings to particular categories of fiction.¹⁶

ALA published a second edition of the Guidelines in 2000.¹⁷ Later, the LC updated instructions in its Subject Cataloging Manual: Subject Headings to aid catalogers in providing increased access to individual works of fiction.¹⁸ As of 2001, the provisions were being applied at the LC to current acquisitions for English-language novels.¹⁹ As a result, while subject headings in catalog records for contemporary fiction (particularly children's literature and popular works that might be found in public libraries) are not uncommon, subject headings for older literary works are much less common. Because general library practice does not require subject headings for literary works for full-level cataloging, one can assume that a large proportion of literature records in academic catalogs lack them, depending on local policies. Although both editions of the Guidelines include genre terms for forms of literature other than fiction, such as poetry and drama, no high-profile projects have focused on providing subject access to literary forms other than fiction, so fewer records for those forms have any subject access.

While the pilot study cited above found the match rate between the academic library catalog and LibraryThing to be low, the match rate for literature was significantly higher, particularly for twentieth- and twenty-first-century literature in English. The aims of the current study were

• to verify the match rate between the library catalog and LibraryThing for twentieth- and twenty-first-century English language literature on a larger sample of records;
• to assess the accuracy and usability of user tags from LibraryThing by comparing LCSH and user tags from LibraryThing for these works; and
• to determine the frequency of user tags for these works in LibraryThing for various literary genres.

A number of studies have compared controlled vocabulary or professionally created metadata with end user tags, descriptions, and folksonomies. Several articles explore user tags and folksonomies for digital resources in the networked environment. Tonkin and colleagues investigated the collaborative aspects of social tagging in a study to determine whether the number and nature of tags assigned varied depending on whether the tagger was tagging only for him- or herself or with the larger user community in mind.²⁰ They found that tag use and tagger motivation vary depending on the internal culture of the particular tagging community. Kipp examined keywords assigned to journal articles by three distinct groups: user, author, and intermediary (e.g., librarian).²¹ The results indicated that differences in keywords assigned by the three groups were influenced by the discrete contexts in which the indexers approach the material they are tagging. Spiteri looked at the linguistic structure of user-assigned tags from three online bookmarking sites: Del.icio.us (now Delicious), Furl (since defunct), and Technorati.²² She found that the vast majority of user tags represented things, as opposed to materials, activities, events, properties, disciplines, or measures.

Hidderley and Rafferty proposed the concept of “democratic indexing.”²³ In 1997, before the advent of social networking and user tags, they argued that different readers approach fiction in different ways and that each reader's interpretations and responses to fiction have validity. In addition, academic response and interpretation of fiction changes over time, and indexing practices should reflect this. In 2007, Hidderley and Rafferty applied this model to folksonomies, demonstrating that user tags for images in Flickr show an array of interpretations of images.²⁴ However, they acknowledged that for precision and recall purposes, folksonomies have limitations without some sort of institutional control.

In studying tagging and folksonomies, several researchers have identified patterns that are consistent over different tagging systems. Munk and Mørk proposed that these patterns follow a power law distribution.²⁵ The power law, also known as Zipf distributions, refers to phenomena where large occurrences are rare and small occurrences are common.²⁶ In the online environments they studied, Munk and Mørk demonstrated that a small number of tags are the most heavily used within a particular system, as compared to a vast number of tags that are used less frequently. As a user population grows, users imitate other users by reusing the most popular tags. They also found that

individual websites are often described with an insufficient number of tags and with very general keywords, because amateurs do not necessarily understand or have not experienced the need for a stringent and precise hierarchization of the information. In this sense, folksonomies often do not yield better search results than using the same keywords in search machines such as Google, because the majority of the keywords used are very general.²⁷

In an investigation of Del.icio.us, Golder and Huberman also found that “users have a strong bias toward using general tags.”²⁸ Regarding the proportion of use of the most popular tags versus those that are less popular, they found that a pattern emerges after the first one hundred or so bookmarks are applied, in which a repertoire of tags becomes constant.

Cultural institutions such as libraries and museums have begun to actively solicit end user contributions to aid in the description of resources. One high-profile example is the LC Flickr pilot project. The LC posted digital images from two photographic collections to the popular photo-sharing site Flickr.²⁹ The project's primary objectives were to increase awareness of the LC's photographic collection, to gain an understanding of the mutual benefits of social tagging and end user input to the LC and the community, and to gain experience in engaging the emergent web community. According to the project report, the public response was “overwhelmingly positive and beneficial.”³⁰As of October 2008, only twenty-five instances of inappropriate (i.e., falling below an acceptable level of civil discourse) user-generated content were found out of 7,166 comments and 67,176 tags. Through the pilot, the LC was able to “collect user-centric, relevant terms that have the potential to increase retrieval of items in the Library's collection.”³¹ At the same time, in sampling the tags supplied for each collection, 45 percent of the tags of one collection and 23 percent of the tags in the other repeated words and phrases already present in the LC-supplied description.

Another experiment in soliciting user contributions to enhance or complement standardized institutional descriptions is Steve: The Museum Social Tagging Project (www.steve.museum).³² Users were asked to provide descriptive tags to digital images of museum objects, to get a nonexpert, noninstitutional perspective. Eighty-six percent of the tags assigned used vocabulary not found in the museum documentation, and 88 percent of these tags were assessed as useful by museum staff. Also speaking to user motivation, Matthews and colleagues conducted an experiment in which users were asked to tag resources in a database of selected web documents.³³ A survey of the participants indicated that some were more likely to be motivated to provide subject terms for general use if there was some indication that the terms would actually be used by and benefit others.

Several articles have been published comparing LCSH with user-supplied keywords or tags. Although different methods were used to determine matching between controlled and uncontrolled terms, the studies universally concluded that the two methods of providing subject access to library resources are complementary, rather than showing that one method is superior. Wetterstrom compared user tags assigned to general collection books from the National Library of New Zealand to the LCSH terms for the same books.³⁴ Terms were coded according to whether they were an exact match to the LCSH heading, a partial match (matching a cross reference or subdivision, or representing a spelling variation), or no match (differences in specificity, point of view, geographic differences in vocabulary, currency of term, or use of more popular language). Seventy-five percent of the tags did not match LCSH, and the highest instances of the divergence from LCSH included use of more popular language, related terms, and broader or narrower terms.

Strader examined the overlap between author-assigned keywords and LCSH for electronic theses and dissertations.³⁵ Matching was analyzed according to whether the term was identical to a main heading or cross reference, whether the same words appeared in a different order, matched partially, or other variations. She found that keywords and LCSH were complementary, and both the controlled and uncontrolled terms provided unique terms that were not otherwise present in the record, aiding retrieval by keyword.

Rolla compared LCSH headings to LibraryThing tags for forty-five books.³⁶ As other studies have indicated, Rolla found that LibraryThing taggers often used broader or narrower terms than supplied by LCSH or used different terminology to identify the same concepts. Additionally, every LibraryThing record included tags for at least one concept not brought out by the LCSH. Rolla also noted that the structure of tag clouds gives more weight to some user tags on the basis of the the aggregation of multiple users’ tags. This is more useful for works with greater numbers of tags because inaccurate or misleading tags are less likely to come to the forefront than in records with few tags.

Weaver conducted an experiment to generate user tags for fiction in a more structured environment than free-form tags.³⁷ Public library users were asked to create tags for the novel The Da Vinci Code using an input form that prompted the taggers to create tags according to specific facets: character, plot, subject, setting, and genre. Although some personal tags were elicited, generally Weaver felt that the resulting folksonomy provided a richer description of the book than did the tag cloud for the same title in LibraryThing.

The results have demonstrated that the number of user tags in LibraryThing is high for the types of twentieth- and twenty-first-century English-language literary works found in a large academic library—the overlap of matching work records between the library catalog and LibraryThing is 82.7 percent. However, the availability of records in LibraryThing for fiction is notably higher than for other literary forms, such as drama and poetry.

The record sampling also showed the lack of subject access for many twenty- and twenty-first-century literary works in academic library catalogs; 61 percent of the catalog records in the sample had no subject headings. Conversely, only 7.1 percent of the 367 work records in LibraryThing had no tags. Records with tags were found in LibraryThing for 70.4 percent of the catalog records lacking subject headings, while another 150 records had subject headings in the catalog as well as LibraryThing records with tags. This high rate of tag availability (76.8 percent of the full sample of 444 records) indicates a strong potential for user tags to enhance the subject access to twentieth- and twenty-first-century English-language literary works in academic library catalogs.

However, a closer look at the LibraryThing tags reduces this potential somewhat. While personal tags are heavily used in LibraryThing, these are eliminated in LibraryThing for Libraries (LTFL), and their prevalence was less problematic than the author anticipated when gathering the data for this study. Only 5 records examined had no usable tags for comparison. What was more discouraging for library purposes was the prevalence of general terms such as “American literature,” “Fiction,” and “Poetry.” This seems to indicate that a lower percentage of the tags than originally thought might be useful to librarians and end users, even when personal tags are eliminated. This is especially true with drama and poetry, where a third or more of all the records examined contained only broad terms, and fewer total tags were assigned per record. Conversely, many records for novels in LibraryThing contained a rich tag cloud of terms. While broad, general tags are often the most popular and heavily used by LibraryThing users, consistent with tag distribution found in other online tagging communities, additional terms bringing out topical and thematic elements are abundant for some works.

A library wishing to utilize end user tags from LibraryThing or another source has several options for incorporating the tags into the library's public catalog interface. A number of next-generation catalog interfaces, such as Innovative Interfaces's Encore or Ex Libris’ Primo, can show tag clouds as part of the record display. Depending on the technical capabilities of the particular discovery software, the tags may be searchable as keywords or usable for browsing. For example, a user viewing a record that displays the tag “Jewish immigrants” might be able to click on that tag to find other library resources on that topic. Next-generation interfaces typically encourage end user interaction, including tagging. While evidence shows that user tagging is more popular in national or multi-institutional sites than in site-specific library catalogs, seeding the catalog with tags from a source like LibraryThing might motivate more catalog users to begin tagging materials in the catalog for their own use or to aid others in finding and identifying materials of interest.³⁹

When tags are used in the discovery layer of a library's catalog, they are not in any way part of the catalog database that underlies the user interface. For several reasons, including accuracy, appropriateness, or privacy, a library may not wish to have user data as part of catalog records, but one disadvantage to including user data is that the data will not easily migrate if the library converts to a new library catalog system. Advances in cloud computing and network-level data sharing make this less of an issue than it might have been in the past, but libraries should be aware that migrating discovery layer data is a separate consideration from converting catalog record data from system to system.

If a library does wish to use tags to enhance catalog record data directly, it might accomplish this in several ways. The MARC field 653 is for an uncontrolled index term. More fully defined, it is intended for an “index term added entry that is not constructed by standard subject heading/thesaurus-building conventions.”⁴⁰ User tags could be imported into 653 fields to provide keyword search access. Another option is the 69X fields, which are “reserved for local subject use and local definition.”⁴¹

Some libraries might be reluctant to incorporate user tags directly into their catalog records, especially if they have concerns about accuracy of terminology or appropriate level of specificity. But for titles that have rich, detailed tag clouds, user tags could aid catalogers in creating LCSH. Tags might identify thematic elements that are not readily apparent from publisher-supplied information on dust jackets or back cover copy, especially because a LibraryThing tagger is more likely to have read the entire book, as opposed to a cataloger downloading or creating a record at the time of a book's receipt by the library.⁴² If an appropriate level of confidence in the accuracy of the tags can be reached, tag terms could be converted to equivalent LCSH terms and placed in 65X fields. In addition, if particular terms seem prevalent among users across a number of titles, a cataloger could consider submitting a proposal to the Subject Authority Cooperative Program (SACO) (www.loc.gov/catdir/pcc/saco) to add the term to LCSH, add it as a cross-reference to an existing heading, or modify an existing heading.

Researchers could further explore a number of areas. For library catalogs that are already using LTFL or other imported tag data, the actual usage of the tags could be investigated. Does the presence of tag clouds increase tagging behavior by users? Do the tags provide keywords not found elsewhere in the catalog record? How does retrieval by tag terms compare with keyword retrieval from other types of record enhancements, such as tables of contents, summaries, or reviews?

Researchers could investigate other social networking sites devoted to books and reading, such as GoodReads (www.goodreads.com) and Shelfari (www.shelfari.com). Hit rates on user supplied data could be compared with library catalogs or LibraryThing to determine relative hit rates and quality of user-supplied content.

While tagging in individual library catalogs has not been overwhelmingly popular, researchers could further explore user motivation. Why have some cultural heritage tagging projects been highly successful, such as Flickr Commons (www.flickr.commons) and Steve: The Museum Social Tagging Project (www.steve.museum)? Something in the presentation of the opportunity for end users to help these cultural institutions has inspired an altruistic spirit in unexpected numbers of members of the general public. Could academic libraries find a way to encourage users to tag by appealing to a feeling of community around the institution's history or traditions? Or is the typical university student or faculty member too focused on his or her own research and concerns to wish to allocate time toward an unproven common good?

Research suggests that fewer library users are starting their search in a local library catalog.⁴³ Many users begin their search with network-level resources, such as Google, Amazon, or WorldCat. How does this affect the need for enriching data in individual catalogs? Should libraries invest instead on centralizing library catalog data and providing links to local information, such as availability?

The purpose of this research was to determine whether user-supplied tags for twentieth- and twenty-first-century literary works in English could enhance or complement the controlled subject headings for the same works in an academic library catalog, and to explore the quality and accuracy of the user tags compared to controlled vocabularies. The author compared a sample of records from a large academic library catalog with corresponding work records in a popular social networking site, LibraryThing. Records were found in LibraryThing for a large number of the works selected (82.7 percent). Nearly 90 percent of fiction was found in LibraryThing, while match rates for other literary forms ranged from 68 to 81 percent. Because of longstanding library practices and traditions, 61 percent of the library catalog records lacked subject headings. Of this subset of works in LibraryThing, 70.5 percent had tags. In addition, 150 works in the full sample had both tags and subject headings.

A comparison of types of tags assigned by end users and library-assigned subject headings found that the most commonly used tags tended to be broader than the controlled vocabulary terms in the catalog records. Very popular or classic works of fiction were more likely to have rich, extensive lists of tags, including many specific terms not found in the subject headings. In many of these cases, the library catalog record and the LibraryThing record provided complementary subject access. Less well-known works, however, and those in literary genres such as poetry or drama, tended to have few if any tags in LibraryThing, and the tags assigned were often very general.

For works that had LibraryThing tags but no LCSH, 42.5 percent of the tags assigned were broad genre terms, and many of the chronologic and geographic tags assigned also were very broad (e.g., “U.S.,” “20th century”). Approximately a third of drama and poetry records had only broad tags, while a majority of fiction works had more specific terms as well as broad terms assigned.

These findings seem to confirm previous studies of end user bookmarking and tagging. General terms may be adequate for personal use or for small, personal book collections. For a large academic library collection, however, broad terms such as “Novel” or “Poems” have less utility in refining searches across many thousands of hits. An academic library wishing to utilize a service such as LTFL might be better served by targeting a smaller subset of records, such as a popular reading collection.

As discovery systems and user behaviors continue to evolve, those working with catalog data must continue to explore ways to enhance access to information resources of all kinds and to improve the user experience. Capitalizing on aggregated end user data about resources is only one way to achieve that aim.