A useful first step is looking at LibraryThing’s explanation of user tags and its instructions on how the site’s members can apply them. In response to the question “What are tags?” the following answer appears on LibraryThing: “The short answer: Tags are a simple way to categorize books according to how you think of them, not how some official librarian does.”¹⁵ Two aspects of this definition are worth noting: the openness of tags—users can categorize their books any way that is useful to them—and the fact that LibraryThing is placing itself in opposition to “official librarians.” The site’s “long answer” continues these two themes:

Once you have a hundred books or so, you need some way to organize them. Library subject classifications, including that of the Library of Congress, are one solution. For most personal libraries, however, they aren’t much use. “Tags,” informal, personal markers used on blogs and sites like Flickr and Del.icio.us, provide a better model.

Here are two examples from my (Tim’s) experience:

1. The LC catalogs Bean’s Aegean Turkey, a guide to the archaeological sites of Turkey’s western coast, under the single subject, “Ionia.” For me, however, the book is about turkey [sic] and archaeology, tags I’ve applied to dozens of books, including Bean’s other archaeological guides.
2. The LC thinks Bernadette Brooten’s Love between women: early Christian responses to female homoeroticism is about six different things, including the mouthful “Bible. N.T. Romans I, 18–32—Criticism, interpretation, etc.—History—Early church, ca. 30–600.” I get by with the tags early church, and homosexuality. To these I added the tag divination. Although the book doesn’t say much about divination, its comments on the topic were actually the reason I picked it up.

Tags can also mark “favorites” or “books to read.” I’ve used the tag ben’s to mark books I should return to my friend Ben. (That I included them in my catalog is, however, a bad sign for that!)¹⁶

These instructions clearly exhort the site’s members to add tags for whatever reason they find useful, even for very personal instances like “books borrowed from Ben” and “books at the summer house.” In addition, both sets of instructions place user tags squarely in competition with LCSH and make the claim that tags are better. The present study aims to explore that claim.

Perhaps the most dramatic difference between the application of user tags and Library-assigned LCSH is that the website’s users assign many more tags to books than library catalogers assign subject headings (see figure 1). Each of the forty-five books under consideration had more user tags in LibraryThing than subject headings in the catalog record by a large margin. The LibraryThing records for these titles had an average of 42.78 tags, while the library records had an average of 3.80 subject headings per record. In LibraryThing records, an average of approximately 7 tags on each record consisted of personal terms; personal terms are explored in depth in the next section. Even disregarding these personal terms, there was an average of 35.16 user tags per record, still much higher than the average of 3.80 LCSH per record.

LibraryThing allows tags of more than one word, and LCSH are made up of strings of pre-coordinated terms, so both tags and subject headings can be broken down into individual keywords. Thus a tag cloud in LibraryThing might include the tags “British empire” and “British history,” which would count as two of the 42.78 average tags but would also count as three keywords, since the word “British” is repeated. Similarly, an LCSH subject string like “Mexico—Emigration and immigration—Social aspects” has five keywords in it, ignoring the conjunction. If keywords (not full tags or subject headings as seen in figure 1) are counted, then LibraryThing still averages more per record: an average of 45.42 per record (37.38, if the personal terms are not included) versus an average of 9.99 keywords represented by LCSH.

Another aspect of LibraryThing tags is immediately noticeable. The user tags in these records contain many personal or individual terms, just as Golder and Huberman remarked in their study.¹⁷ Tonkin and colleagues found that personal terms, which they call “time, task, or project labels,” form 16 percent of unique tags on Delicious.¹⁸ LibraryThing’s instructions for tagging books encourage the use of personal terms. In addition to the examples of “Ben’s books” and “summer house” found in the website’s instructions, terms like “book group,” “book club,” “read,” “unread,” “to read,” “own,” “read in 2007,” and “not in library” frequently appear in the records studied. Several tags of this type appear in each of the forty-five records studied here. These terms may have strong personal value to the users who input them but, from a library’s perspective, they are not useful descriptor terms and do not provide any subject access to the books. LibraryThing’s tag cloud display, like most such displays, does give weight to more popular terms by increasing the size of words in the display according to how many users have assigned those tags to a work. These personal or individual terms are usually very small, indicating that only a few people have added that tag to a record, so that visually, at least, these personal terms do not predominate in the tag cloud display. If libraries permit user tags in their catalogs, they will have to decide whether to allow and how to handle these individual and perhaps unhelpful tags. Conversely, certain personal tags (i.e., those that do not directly relate to the contents of the book but to the user’s experience with the book) could have practical value in a library catalog environment. For example, if students or professors come across books that are useful for a particular course, they could tag the book’s record with the course number to help other students find the same book.

In every LibraryThing record, the user tags contained at least one concept not covered by the subject headings in the catalog record. In many cases, these concepts represented ideas that a cataloger would not have brought out, deeming them irrelevant to the overall content of the book or somehow not consistent with the typical practice of subject analysis for books. Comparing LibraryThing’s instructions on assigning tags with the LC’s instructions to catalogers on how to assign subject headings illuminates some of the differences between the use of tags and subject headings. In section H180 of the Subject Cataloging Manuals, the LC instructs catalogers to “assign to the work being cataloged one or more subject headings that best summarize the overall contents of the work and provide access to its most important topics.”¹⁹ Compare this to the longer instructions given by LibraryThing, in which a reader assigned the tag “divination” to a work not because the book was primarily about that topic but because its comments on divination interested him. The fact that the users of LibraryThing assign tags to books representing concepts not brought out by LCSH does indicate that catalogers, by following the LC guidelines, may omit concepts that are important to users.

For each of the forty-five titles in this sample, the LibraryThing tags contained subject terms or concepts that the subject headings did not express. That figure does not include the personal or individual terms, but words and phrases describing the subject of the book. Conversely, the librarian-assigned subject headings in twenty-five records (55.6 percent) brought out concepts and topics that the user tags did not. Finally, the subject headings and user tags assigned to thirty-five records (75.6 percent) brought out the same subject or concept, although often expressed in different terms. Thus, approximately three-quarters of the time, catalogers and readers agree on at least part of what each book is about, even if the tags and subject headings express the content of the book differently. The specific ways in which the user tags and subject headings for these titles differ is instructive, and these differences do suggest that adding user tags to library catalogs could help improve subject access to collections.

First, user tags almost always include very general and broad subject terms. In each of the forty-five records under consideration, LibraryThing’s users added general or broader terms for the concepts discussed. Books about Africa show this clearly. Several of the books under consideration discuss the genocide in Rwanda, and at least one deals with the civil war in the Congo. The catalogers, following the instructions for assigning LCSH, assigned headings relevant to the specific events and countries. For example, the subject heading string “Rwanda—History—Civil War, 1990–1993” appears in a bibliographic record. The LibraryThing records for books on Rwanda and the Congo, however, all contained tags such as “Africa,” “African history,” and other similar broader terms. Catalogers did not—and as a rule do not—include broader geographic terms. The books on Mexican immigrants show the same pattern. Catalogers used headings that included the term “Mexican American,” but LibraryThing’s users added terms like “Hispanic” and “Latino,” which represent the broader population of which Mexican Americans are a part. Other examples from LibraryThing include the tags “science,” “modern history,” or “world history.” Catalogers would only use a term like “science” for a book that is an introductory textbook or that discusses the entire discipline of science. Similarly, a book with a cataloger-supplied subject heading “World history” would have to discuss the entire history of the world. Many LibraryThing users, however, following the pattern discussed by Munk and Mørk as well as Golder and Huberman, assign these broader subject terms to works that discuss a specific discipline in science or a specific time period of history.²⁰

Conversely, LibraryThing users often add terms that are more specific in nature than the subject headings supplied by catalogers. Eleven of the forty-five records (24.4 percent) examined in LibraryThing contained narrower or more specific terms than the librarian-supplied subject headings. These specific terms frequently described books that were general in nature or subject matter, such as comprehensive histories of Africa. For these books, the library catalogers typically followed standard practice and only assigned a broad heading appropriate to the overall content of the book, such as “Africa—History.” LibraryThing’s users, however, added terms for more specific concepts, such as “slavery,” “colonialism,” and “exploration.” The records for Stephen J. Hawking’s popular A Brief History of Time also show this pattern. The record in the library catalog contains a single and very general subject heading, “Cosmology.” LibraryThing users, however, supplied several more specific tags, such as “physics,” “astrophysics,” “big bang,” and “black holes.”

One specific example shows all of these differences between LibraryThing tags and library-supplied subject headings. The WorldCat bibliographic record (see figure 3) for the book A Savage War of Peace: Algeria, 1954–1962 has one LCSH, “Algeria—History—Revolution, 1954–1962.” In LibraryThing, following the expected pattern, users have assigned many more tags to the book (see figure 4). Some tags express the same concept as the LCSH but in different words, like “Algerian history,” “Algerian Revolution,” and “Algerian War.” In addition, many LibraryThing users have added tags that describe much broader concepts or that refer to broader geographic areas, like “20th Century,” “Africa,” “Middle East,” “North Africa,” “Military history,” and “war.” Other LibraryThing tags also bring out more specific concepts that may not relate to the content of the book as a whole but that some users found important, such as “colonialism,” “guerrilla,” “counterinsurgency,” and “torture.” Interestingly, in this LibraryThing record, more users assigned the tags “France” and “French history” than the tags “Algeria” and “Algerian history,” whereas the subject heading that relates to the particular war described in this book does not even mention France.

Wetterstrom’s results in the study comparing user-assigned tags and LCSH differ from results reported here.²¹ As mentioned above, Wetterstrom asked twenty people to contribute tags to a small collection of books. Wetterstrom’s study group assigned significantly fewer tags than found in this study’s LibraryThing records. Wetterstrom found an average of 24.4 user tags compared to the average of 42.78 in the LibraryThing sample. Wetterstrom’s results also differ from those seen in the present comparison of tags and LCSH. In his study, 75.47 percent of tags did not match LCSH, fewer matches than seen here. Wetterstrom’s study also found a preponderance of broader and narrower terms in the tags (14.61 percent and 19.62 percent of overall tags, respectively), although he only counted the overall number of these broader and narrower terms and did not specify if terms of these types were assigned to every book. In Wetterstrom’s study, narrower terms appeared more frequently than broader terms, which is the opposite of the results found in this study. The difference in findings may be because of two reasons. First, Wetterstrom’s study group was consciously creating tags for a research project and not for describing books in their own personal collections and which they wanted to be able to retrieve. In other words, the LibraryThing users had a personal investment in the creation of the tags and were not part of a research study group. Second, Wetterstrom’s study group (twenty individuals) was much smaller than the universe of people contributing tags to LibraryThing, who could view tags added by others and who, in total, created significantly more tags.

Examining the LCSH assigned to the forty-five books is instructive. Twenty-five of the WorldCat records (55.6 percent) contained LCSH expressing concepts that LibraryThing’s tags did not. These differences fall into several categories. First, LCSH in these records often refer to classes of persons, while the user tags generally only indicate abstract concepts. For example, books about Mexican immigrants have LCSH such as “immigrants” and “Mexican Americans,” nouns that refer to the groups of people. The user tags in LibraryThing for these same books, on the other hand, are “immigration” or “Mexican American,” “Chicano,” and “Latino,” that is, either the abstract concept or adjectives rather than nouns.

LCSH contains a collection of phrases known as “free-floating subdivisions.” Catalogers can append these free-floating subdivisions, with certain restrictions, to topical headings that already exist in LCSH and thereby highlight special aspects of the topical heading. Nothing like these free-floating subdivisions appears in LibraryThing’s user tags, and some records for the forty-five books considered here demonstrate the usefulness of these subheadings. Free-floating subdivisions are standardized phrases and are often not expressed in natural language. A common example of a free-floating subdivision is the term “Social conditions,” which catalogers can add to the names of places or to a phrase denoting a class of persons. The LC’s Subject Cataloging Manual defines this term in the following manner: “Use the subdivision for works discussing the social history or sociology of a place, ethnic group, or class of persons, including such subtopics of sociology as social problems, stability, change, interaction, adjustment, structure, social institutions, etc.”²² The term “social conditions” does not appear as a user tag, and one can see LibraryThing’s users struggling to come up with a good way to convey what the subdivision “Social conditions” expresses. Tags like “social problems,” “social history,” and “sociology” appear, but are not usually the bold-faced, more popular terms, so LibraryThing users do not seem to have a clear consensus on how to express this concept.

The tags in LibraryThing also fail to show any consensus with the expression of historical time periods, whereas in LCSH chronological divisions are established for all countries and regions. These chronological subdivisions vary from country to country and, ideally, conform to the important historical divisions within each country’s history. The library bibliographic record for A Savage War of Peace, as shown above, has the subject heading “Algeria—History—Revolution, 1954–1962.” The LibraryThing records for this book had very few chronological tags, and the few tags in the record relating to time periods (“1950s,” “1960s,” “20th century,” “pre–1983,” and “post–war”) lacked specificity and consensus. The chronological subdivisions in the LCSH thesaurus do require time and effort on the part of catalogers to establish as well as to apply correctly. As the LibraryThing tags show, users do not necessarily think about historical events in neat chronological packages. The chronological subdivisions, however, like the free-floating subdivisions, do serve the purpose of bringing together materials about a given subject or, in this case, a given time period.

The LCSH system, like all thesauri and taxonomies, controls synonyms and also has an elaborate set of rules for how new subject headings are established. These rules include such basic grammatical guidelines as which types of nouns should be in the plural in subject headings and which in the singular. A subject heading that appeared frequently in this study, for example, is “Mexican Americans,” for which the LCSH thesaurus lists two nonpreferred terms, “Chicanos” and “Hispanos.” Within this system, catalogers know which term to add to a bibliographic record, and users, once they know which is the preferred term, theoretically can find all the materials under that topic. User tags in LibraryThing, however, like the tags and folksonomies studied by the authors cited above, neither control synonyms nor follow specific grammatical rules. As a result, various terms that mean more or less the same thing but are expressed differently (“Mexico,” “Mexican,” “Mexican American,” “Chicano,” and “Latino”) appear in the same LibraryThing record. This situation occurred in each of the forty-five sample records. In each record several different subject terms, or terms in different grammatical forms, expressed the same concept. Other examples include “current affairs” and “current events”; “diet,” “eating,” and “nutrition”; “Jewish,” “Jews,” and “Judaism”; “economics” and “economy”; “mountain climbing” and “mountaineering”; “decision-making” and “decisions”; and “thinking” and “thought.” Both noun and adjectival forms were used for geographic descriptors, for example, “Africa” and “African”; “Mexico” and “Mexican”; “Iran” and “Iranian.” Librarians who defend LCSH, including Mann at the LC, base their defense on the ability of subject headings to control synonyms and specify the correct grammatical form to use, which helps to collocate similar materials.²³ With all the variations possible in LibraryThing, users searching for a specific topic cannot be sure that they have found all the relevant books because they cannot necessarily predict the terms or grammatical forms that other readers have used. LibraryThing has recently added a new feature, in which users can make two tags equivalent to each other, to try to control synonyms. When that happens, only the more popular tag displays.²⁴ This interesting innovation can help reduce the redundancies in the tag cloud; however, combining tags only works when they are identical in meaning and use.²⁵ Tags like “economics” and “economy,” for example, cannot be combined. Controlled vocabularies like LCSH, in addition to choosing preferred terms from synonyms, can also provide scope notes on how to choose between two similar but distinct terms.

Controlled vocabularies, then, help catalogers choose the appropriate subject headings to use. LibraryThing’s users, however, have an advantage over catalogers when they assign tags—they have probably read the book before tagging it on the website. Catalogers do not have the time to read an entire book before assigning subject headings and, therefore, base their subject analysis of books on words in the title, publishers’ blurbs, the preface of the book, and chapter titles. Catalogers can also be constrained by the fact that they are trying to assign subject headings that will be meaningful for a large group of unknown and potentially diverse end users, and they may not know what subjects in a book will be most important to all potential readers. In LibraryThing, the cataloger and the end user are the same.

User tags, as Spiteri has pointed out, also can adapt better and more quickly to changing terminologies and to new fields of study than LCSH or any controlled vocabulary can.²⁶ The LC has to approve new terms added to LCSH, and catalogers have to do a certain amount of research before proposing a new topical heading.²⁷ Because of this, new headings take time to appear in LCSH. In addition, LCSH are formulated to avoid polemical topics and maintain an objective stance toward the material, which often has the reverse effect of indicating a subtle bias. In the example of the book A Savage War of Peace, the only LCSH in the bibliographic record is “Algeria—History—Revolution, 1954–1962.” This subject heading does not explicitly mention France and its involvement in the war, while conversely more of LibraryThing’s tags for this book cited France than Algeria. The library-supplied subject heading, then, subtly erases the anticolonial nature of the war. The political implications and biases of the language used in LCSH have long interested researchers and have also inspired projects such as Berman’s alternative subject headings.²⁸ The present study is more concerned, however, with the issue of subject access to materials. A solitary subject heading like “Algeria—History—Revolution, 1954–1962” potentially hinders access to the book for readers because, according to LibraryThing tags, this book interests many users because of what it says about French history. Users browsing or searching for subjects in the library catalog for books about French military history will not come across A Savage War of Peace.

The subject headings in library catalog records also can suffer from what can only be called bad cataloging. Several of the books considered here have very inadequate subject headings in the library record. The book Fast Food Nation by Eric Schlosser provides an especially egregious example. Because this is a widely read and widely discussed book, many people, even if they have not read it, know that it is an indictment of the fast food industry that discusses the environmental, economic, and nutritional aspects of fast food restaurants. In the library catalog, though, this book’s record only has one subject heading: “Cookery, American.” Even leaving aside the unnatural language in this subject heading (one example of the LC’s reluctance to change headings once established), this subject heading does not accurately describe the contents of the book. It makes the book sound like a cookbook and not a social and political look at a specific part of the American food industry.

In LibraryThing, since more than one person can add tags to a record, the overall tag cloud for the book can correct a single person’s errors or questionable judgment. This does point to a potential disadvantage. If only a few users have added tags, then the aggregate of tags assigned may not provide the most accurate or helpful subject analysis of the book. A few of the records under consideration here had tags supplied by only six or seven readers, and in those instances the tag clouds were not comprehensive and the tags did not provide a good description of the contents of the books. The better and more complete records in LibraryThing usually belong to more popular books, which does have certain implications for user tags in library catalogs. A large portion of the collections of university and research libraries consists of rare and little-known materials, and these are often the most useful and valuable items these libraries own. Libraries cannot rely on their users to supply the subject access for rare materials if no users, or even just a very few, have read the books. Also, the university community consists of professors and students doing higher-level research who often will need to find every book on a certain topic. To serve these scholars, libraries need to provide consistent and comprehensive subject access, and also need to use some means of controlled vocabulary that will bring together all similar materials.