Representation of Swedish LGBTQ+ fiction in commercial information services

When searching for LGBTQ+ fiction in Sweden, end users typically consult social media. For obtaining a book where the author or title is already known, they will then use the library catalogue (Golub et al., 2022). Typical search problems they face in different information search services are related to too few LGBTQ+ books being indexed with LGBTQ+ keywords. In addition, subject indexing systems in libraries tend to be too shallow or too general (Golub et al., 2022). To address this problem, the Queerlit project (2021–2023) created a dedicated database of Swedish LGBTQ+ fiction (https://queerlit.dh.gu.se/), implementing deep and specific subject indexing based on the Swedish QLIT thesaurus for LGBTQ+ themes.

Still, we can assume that many Swedish LGBTQ+ fiction readers will continue to use their established information channels like social media and web search engines such as Google to identify works of interest. At the same time, metadata on these information channels may provide a source of complementary subject index terms that could be added to the information professionals’ indexing. Therefore, this paper aims to explore: (1) to what degree are Queerlit books identified as LGBTQ+ books in widely used information services, in particular Google Books, LibraryThing and Goodreads; and, (2) whether metadata in these information services could be considered of value for the Queerlit database and complement its highly extensive and highly specific indexing.

The paper is structured as follows. The Background section provides a brief overview of challenges pertaining to finding LGBTQ+ fiction and the Queerlit approach is outlined. The section also examines the indexing policies of Google Books, Goodreads and LibraryThing. The Methodology section concerns the sample, data collection and analysis techniques. The coverage of LGBTQ+ works from Queerlit in Google Books, Goodreads and LibraryThing is presented and discussed in the Results section, which also addresses the degree to which LGBTQ+ works are identified as such in the information services. This section also includes examples of tag clouds and an in-depth analysis of social tagging challenges, exploring their potential as complements to professional subject indexing in library catalogues. In the Conclusion, the main findings are summarised and recommendations for future research are given.

Queer literature scholars stress the significance of fiction in LGBTQ+ history (see, e.g. Love, 2007; Heede, 2015; Wallace, 2016), yet users face challenges finding LGBTQ+ fiction (Golub et al., 2022). To some degree this is related to fiction indexing in libraries being limited to genre, time and place (Saarti, 1999), while users are often looking for aspects such as storyline, characterization, tone etc. Specific to LGBTQ+ indexing, a common problem is that general controlled vocabularies tend to include only broad LGBTQ+ terms, making distinction between different LGBTQ+ works harder to discern for users. Challenges also arise in indexing the subjective themes, especially when intentionally hidden, as seen in pre-1990s fiction with implicit references to homosexuality.

Social tags might complement professional subject indexing, as noted by Rafferty (2018), Rolla (2009), Kipp and Campbell (2010) and Golub et al. (2014). However, social tagging suffers from drawbacks such as the absence of consistent indexing rules, users employing various tags denoting the same concept, unresolved homonymous tags. Furthermore, social tags often lack a hierarchical relationship between tags, include variations in tag forms (e.g. singular/plural and spellings) and can include an unlimited number of tags with relevance limited to personal use (e.g. “to read”) (Furner, 2010).

Complementing subject indexing in libraries with folksonomies (tag clouds) could be of particular interest for fiction indexing as themes other than genre, time and place are captured by social tag clouds (see, e.g. Johansson and Golub, 2019). This could also be the case for LGBTQ+ works (Adler, 2009; Bates and Rowley, 2011). Adler (2009) studied how 20 books on transgender topics are described in WorldCat using Library of Congress Subject Headings and in LibraryThing with social tags, showing that relevant terms frequently used by users are not as prevalent or sometimes completely absent in WorldCat. Additionally, LibraryThing allowed for a more varied range of gender expressions, which are often lacking in Library of Congress Subject Headings. Bates and Rowely (2011) also showed that LibraryThing folksonomy offers advantages when it comes to discoverability and representation of LGBTQ+ materials, but can also reflect biases. For instance, the overrepresentation of US taggers on LibraryThing impacted the tagging of ethnic minority resources.

The Queerlit database incorporates first editions of individual printed works of fiction originally published entirely or partly in Swedish, in one of the national minority languages within the current borders of Sweden, as well as in Sami or in Meänkieli when published within the borders of Scandinavia. Furthermore, the database encompasses works translated into Swedish and those authored by individuals primarily active within Sweden’s current borders. Autobiographical texts, including essays, letters and diaries, which extensively address the LGBTQ+ themes, are also included.

Subject indexing of LGBTQ+ themes in the Queerlit database is based on a dedicated QLIT thesaurus for LGBTQ+ terms. The thesaurus incorporates relevant terms from the international Homosaurus, translated to Swedish and adapted for fiction, guided by input from users, including librarians and subject experts. For non-LGBTQ+ terms, the following vocabularies are used:

1. Swedish Subject Headings (Svenska ämnesord (SAO)), for general themes in adult fiction;

2. Children’s Subject Headings (Barnämnesordslistan (Barn)), for general themes in children’s and young adults’ fiction;

3. “SAO” for Genre and Form (SAOGF), for genre and form in adult fiction;

4. “Barn” for Genre and form (Barngf), for genre and form in children’s and young adults’ fiction.

In addition to having its own search interface and thesaurus, the Queerlit database is a sub-database of Libris, the Swedish union catalogue. The Queerlit indexing policy is aligned with the Libris policy which prescribes adding a subject if at least 20% of the work is about that subject (Queerlit, 2023a). However, for works of fiction this rule is more relaxed to include a subject or theme that stands out (Queerlit, 2023a).

When it comes to indexing a work of fiction with an LGBTQ+ term, the Queerlit database aims to be as inclusive as possible. Therefore, the database includes works that either explicitly address LGBTQ+ themes or works where these emerge from the context identified by researchers or LGBTQ+ communities (Queerlit, 2023b). Queerlit thus provides both major and minor, or peripheral index, terms.

The indexing policy further advises using the most specific subject term available, which is a common indexing principle in information search services curated by information professionals. The database has four librarians who review each other’s metadata and discuss subject indexing on a continuous basis to help increase inter-indexer consistency. In addition, the QLIT thesaurus has a scope note for each term, describing how the term should be used, which contributes to higher inter-indexing consistency.

Figure 1 shows an example of a metadata record for the book titled “Route 66 går till Trollhättan” by Bodil Sjöström from 2011 (direct link to the record: https://queerlit.dh.gu.se/work/zh9bxqn95d9s34q/route-66-gar-till-trollhattan). The book has seven subject index terms from QLIT, of which one is a major theme, denoted by the colour of dark orange and larger font: lesbian couples (in Swedish: lesbiska par). There are also six peripheral LGBTQ+ themes, denoted by light orange and smaller font: civil unions (registrerade partnerskap), lesbophobia (lesbofobi), tomboys (pojkflickor), coming out (komma ut), slurs (skällsord), HIV/AIDS (hiv/aids). It also has four SAO terms, in the colour grey, which are non-LGBTQ+ terms: car trips (bilresor), United States (Förenta staterna), families (familjer), Alzheimer’s disease (Alzheimers sjukdom). SAO genre terms are novels (romaner) and bildungsroman (utvecklingsromaner). There is also a classification notation (klassifikation), a Dewey Decimal Classification (DDC), 839.738, taken from Libris, which represents Swedish novels and short stories published from 2000 onwards (Svenska romaner och noveller – 2000-).

Google Books (https://books.google.com/) is a Google index for full-text books, launched in 2004 as Google Print (“Google Books”, 2023) which today offers access to searchable text of millions of books (Chalmers and Edwards, 2017). The books are digitised in collaboration with libraries (this program is called the “Library Project”) as well as with publishers and authors (“Partner Program”) (Google, 2024).

According to Google (Google (2024), metadata for the books are taken from various sources: the publishers participating in the program, books scanned through the Library Project, metadata providers that aggregate book metadata like EBSCO and The National Diet Library, Japan, as well as Wikipedia and related websites. No other information is provided related to indexing on this website. It is stated that Book Search works like web search: “Try a search on Google Books or on Google.com. When we find a book with content that contains a match for your search terms, we’ll link to it in your search results” (Google (2024)). This indicates that assigned indexing – using a controlled vocabulary – is not implemented in any systematic way.

A key implication of automatic metadata aggregation for (LGBTQ+) fiction searching is its suitability for known-item searches, where the user already knows the author and title. However, earlier research has highlighted significant issues with metadata accuracy. Nunberg (2009) identified problems with misclassification, while James and Weiss (2012) reported a high error rate of 36.75% across four major metadata elements: author, title, publisher and publication date. Although Chalmers and Edwards (2017) later observed substantial improvements in Google’s automatic metadata aggregation, they noted that “the metadata continues to be far from accurate”.

Goodreads is a social cataloguing service, launched in 2007 and acquired by Amazon in 2013. In 2019 it had 90 million members (Goodreads, 2024a, b) and in 2024 it lists 3.5 billion books (https://help.goodreads.com/s/question/0D58V00008Rm69PSAR/how-many-books-are-listed-on-the-goodreads-site). Metadata were added over the years from different providers, including The Library of Congress, WorldCat, Amazon, Ingram and individual publishers (Goodreads, 2024a, b). Goodreads Librarians Group (Goodreads, 2024a, b) are volunteers who help ensure accuracy of the metadata, but this does not seem to include end user tags. Goodreads creates the metadata element “genre” by crowdsourcing end user tags (that it calls “shelves”) (Goodreads, 2019). There appear to be no guidelines to ensure tagging consistency, such as pre-existing terms to choose from or regulations addressing relationships between synonyms, homonyms, different word forms. Similarly, there is no framework for managing broader, narrower or related terms.

LibraryThing (https://www.librarything.com/) is another social cataloguing service, launched in 2005. As of March 2024 it had 2.9 million users (https://www.librarything.com/zeitgeist) and in 2021 it listed 155 million books (https://en.wikipedia.org/wiki/LibraryThing). In addition to serving end users, it provides services for libraries, including “LibraryThing for Libraries (LTFL)” which enables import of metadata elements such as LibraryThing tags and book recommendations into library catalogues.

LibraryThing takes metadata from Amazon and libraries via the Z39.50 protocol (https://en.wikipedia.org/wiki/LibraryThing). For tagging there is a limited degree of vocabulary control: in the process it calls aliasing, the LibraryThing links synonyms and selects the most popular synonym as the final, preferred form. For example, “New York” and “New York” have been linked to one single tag ID, with “New York” chosen as the preferred term (http://librarything.com/services/feeds/tags). At the same time, the users are advised that three different capitalisations are considered as three separate tags (Goodreads, 2019). LibraryThing also conducts tag approval and rejection. Furthermore, there is a tag approval process whereby tags that are used a number of times and are considered useful for searching are approved. Personal tags, purely evaluative tags (“great”), tags about a specific edition (“audio book”), obscene tags and ambiguous tags are rejected. The end-user interface seems to show the approved tags only and this is what we have also used in our data analysis.

Like Goodreads, LibraryThing generates the “genre” metadata element. This is based not only on tags but also on library and bookstore classifications and subject headings, such as Dewey Decimal Classification and Library of Congress Subject Headings. LibraryThing also specifically highlights that genres are not a replacement for tags, but rather a way of grouping similar works (https://www.librarything.com/topic/333142).

The sample consisted of metadata for all works in the Queerlit database as of 17 June 2023 (n = 1,320). The metadata was then automatically searched in Google Books, LibraryThing and Goodreads using Queerlit ISBNs and the following techniques:

1. Google Books: via the Google API using ISBN numbers from Queerlit metadata and a collection script in Google Sheets, visiting from a Swedish IP address. This was conducted during the period from 17 to 25 June 2023. The script is available in  Appendix.

2. Goodreads: the Queerlit ISBN numbers were used to match the Queerlit sample against the Goodreads metadata records. This was done during the period from 17 to 25 June 2023. The matching Goodreads records were then collected using the freely available version of the Octoparse tool (www.octoparse.com) to collect tags from the user interface. The Octoparse tool was applied during July 2023.

3. LibraryThing records were collected using the same process as for Goodreads, for the ISBNs found in Goodreads. This was done in November 2023. The reason for expanding the initial research plan from Google Books and Goodreads to also include LibraryThing was that we wanted to compare metadata on the two social tagging services. However, as the analysis progressed, we saw few major differences between the two services in relation to our research aims. At the same time, the number of the records found was small (n = 100), which is why we have decided to retrieve yet another set from LibraryThing in April 2024, by extracting LibraryThing records for the full Queerlit sample of 1,320 books.

   NB. It is important to note that many titles were added to LibraryThing when the ISBN import for this project was performed. For our purposes, we have excluded all titles with only one user directly after the import, as that user would be us having added the book when performing the import.

Data analysis was conducted: (1) quantitatively, by counting matches between Queerlit and each of the three databases based on string matching the titles; and, (2) qualitatively, by analysing selected metadata records from the three databases manually to explore any challenges that available index terms might cause for users of those services as well as examine the potential added value for the Queerlit database. The qualitative analysis included consultation with the Queerlit indexers.

Out of the 1,320 books from the Queerlit database, only 109 were found on Google Books in June of 2023. Of those 109 books, less than 10% are identified as LGBTQ+ works (10 or 9.17%). Furthermore, a comparison of LGBTQ+ terms from Queerlit to those of Google Books shows that Google Books index terms are much less specific and exhaustive. The only LGBTQ+ terms used by Google Books are general and most refer to identities: “gay men”, “lesbians”, “gays”, “lesbian teenagers”, “bisexuals”. There is only one other LGBTQ+ term used, “coming out”, which turned out to be wrong for the work (see below). Using a low level of indexing specificity and exhaustivity leads to high recall with low precision, and is likely to result in many search results which the user has no option of further narrowing down.

Further comparison shows that two of these 10 books have wrong index terms, which could lead to wrong retrieval results (false positives).

1. For the book titled “Jenny” by Jonas Gardell, Google Books lists “Gay men -- Fiction -- Sweden”. The book is written by an author who openly identifies as gay, but the story revolves around a girl who was raped during a party in high school. In other words, the book is not about gay men. In the Queerlit database the book has only one LGBTQ+ term, and it is a minor one, “transgressing binary gender norms” (könsöverskridanden).

2. For the book titled “Rami och Jonathan” by Mats Berggren, Google Books lists “coming out” as a term. Major QLIT terms assigned to this work in Queerlit are: “gay youth” (unga bögar), “LGBTQ+ love” (kärlek (hbtqi)), “LGBTQ+ athletes” (idrottare (hbtqi)). Minor QLIT terms are: “sex (act)” (sex), “LGBTQ+ death and dying” (döden och döende (hbtqi)), “LGBTQ+ sorrow” (sorg (hbtqi)).

All the 109 books found on Google Books have the metadata element “subject” while 55 also have the metadata element “genre” with values such as “fiction”, “biography”, “thriller”, “mystery”, “science fiction” etc. Inconsistencies in genre terms are evident, as the concept of fiction includes subtypes such as thrillers, mysteries and science fiction. Additionally, different terms are used to label fiction, such as “fiction” and “fiction/general.” This likely reflects earlier observations that metadata is automatically sourced from multiple resources without being checked for consistency or quality.

The subject metadata elements seem to be characterised by similar problems. More than half of the books in the sample have only genre values in the subject (60 or 55.05%). Examples include “literary historians – fiction – Sweden”, “Swedish fiction” or “Children’s literature, Swedish”. Also, the genre values are varied, just like genre values in the genre metadata element.

Of the 1,320 Queerlit books, only 13.3% (n = 175) were found on Goodreads in June of 2023. The service is primarily dominated by English-language works. As mentioned earlier, Goodreads crowdsources user shelves to assign final sets of keywords to the books, which it calls “genre”. Of the 175 Queerlit books found on Goodreads, only 62 had any genres assigned. This suggests that relatively few Goodreads users were familiar with these books. It may also indicate that Swedish-language readers in general, and Swedish LGBTQ+ fiction readers in particular, are less likely to use Goodreads.

Of the Queerlit books with index terms on Goodreads, 25.81% (n = 16) have an LGBTQ+ related term among the genre terms presented as the main tag cloud. Of the remaining 46, 30 have an LGBTQ+ tag among the less rare tags found under the “More” link whereas 16 have no LGBTQ+ terms listed anywhere.

There are several illustrative examples from the group of 30 works with an LGBTQ+ theme outside the main tag cloud. “En komikers uppväxt” by Jonas Gardell had the following terms on the main page: “Fiction”, “Swedish Literature”, “Sweden”, “Contemporary”, “Read for School”, “Scandinavian Literature”; and the following two LGBTQ+ terms under the “More” link: “lgbtq”, assigned by four users, and “hbtq” by two users. Also, “Den ärliga bedragaren” by Tove Jansson had the following terms on the main book page: Fiction, Scandinavian Literature, Finnish Literature, Literary Fiction, Classics, Novels, Swedish Literature; as well as “queer” at the “More” link, added by six users. The author is Finnish and the described book is a Swedish translation, but the combined tags “Finnish Literature” and “Swedish Literature” are confusing to the reader; and so is the term “Scandinavian literature” which excludes Finland. The book “Drömfakulteten” by Sara Stridsberg, has as the following terms on the main book’s page: “Fiction”, “Feminism”, “Swedish Literature”, “Novels”, “Historical Fiction”, “Sweden”, “Contemporary” with additional “More” tags denoting LGBTQ+ themes: “LGBTQ” by nine users, “LGBT” by six and “queer” by five. These examples present indexing errors also when it comes to the established indexing guidelines where only the most specific index term should be added – “Fiction”, “Literary Fiction” – a narrower term of “Fiction”.

Below are three examples that have an LGBTQ+ term in the main tag cloud.

1. “Stjärnor utan svindel” by Louise Boije af Gennäs (https://www.goodreads.com/book/show/1696315.Stj_rnor_utan_svindel) with the following LGBTQ+ terms on the main book page: “LGBT”, “Queer”, “Gay”, “Lesbian” and additional ones under the “More” link: “lgbt” (7), “queer” (7), “lgbt” (“hbtq”, 6), “lgbtq” (5), “gay” (3), “lgbtqiap” (2), “lesbian” (2), “hbtqiap” (1), “queer-fiction” (1), “gay-shelf” (1), “lesbian books” (“lebb böcker”, 1).

2. “Rent spel” by Tove Jansson (https://www.goodreads.com/book/show/25123651-rent-spel) with the main book page LGBTQ+ terms: LGBT, Queer, Lesbian and additional ones under the “More” link: “lgbtq” (54), “lgbtqia” (28), “sapphic” (19), “gay” (6).

3. “Min bror och hans bror” by Håkan Lindquist (https://www.goodreads.com/book/show/2245962.Min_bror_och_hans_bror) with the main book page LGBTQ+ terms: LGBT, Queer, Gay, Gay Fiction; and additional ones under the “More” link: “lgbt” (24), “lgbtq” (10), “queer” (6), “gay-fiction” (6), “glbt” (5), “lgbtqia” (4), “mlm” (2), “m-lgbt” (2), “lgbt-fiction” (2), “queer-lit” (2), “hbtq” (2), “m-m-romance” (2), “lqbtqia” (1), “queer-leads” (1), “lgbt-tbr” (1), “lgbtqi” (1), “the-moon-is-trans” (1), “literary-fiction-queer” (1), “lgbtiaq” (1), “0-gay” (1), “bl_gay” (1), “tbr-lgbtq” (1).

The terms on the main book pages are problematic in that they are conceptual duplicates or synonyms and overlap in meaning. For example, “LGBTQ” and “Queer” are synonyms and only one term should be used throughout the metadata, with links from the non-preferred term at the time of indexing and searching. “Gay” and “Lesbian” are more specific than “LGBTQ” and the principle of assigning the most specific term has not been followed. Also “Gay” and “Gay Fiction” present mixing of genre and subject; “Fiction” is already listed as a term in the fifth example so there is no need to write “Gay Fiction” if there is both “Gay” and “Fiction” already.

These examples also show the known problems of social tagging, outlined in the Background section above, where different terms are used to refer to the same concept (“lgbt”, “lgbtq”, “queer”, “glbt”, “lgbtqia”). There is no word-form control in tagging, resulting in a mix of facets for both subject and genre (e.g. “gay-fiction,” “lgbt-fiction,” “literary-fiction-queer”), as well as tags combining subject and personal use (e.g. “lgbt-tbr”). Additionally, there are no consistent indexing policies, and some tags are exclusively for personal use (e.g. “tbr”). These inconsistencies make information retrieval challenging, requiring searchers to consider all possible synonyms and word forms to retrieve relevant results. At the same time, the terms suffer from the lack of specificity – usually only most general terms denoting LGBTQ+ identities are used.

Out of 1,320 Queerlit books a total of 55.3% (n = 730) were found in LibraryThing. Of those, 36.7% (n = 484) had any tags assigned. This is a much larger representation of Swedish LGBTQ+ books than in Goodreads, in spite of the fact that Goodreads has many more books in their service. It is hard, however, to understand the reasons for this without a study of the two services regarding their coverage, acquisition policies and their user base. Such an analysis being outside the scope of this study, we were at least able to notice that Scandinavian LGBTQ+ organizations such as the Swedish Federation for Lesbian, Gay, Bisexual and Transgender Rights (RFSL) Uppsala (https://uppsala.rfsl.se/) and the Danish LGBTQ Library (https://lgbt.dk/bibliotek/) were common contributors on LibraryThing.

However, like in Goodreads, only a small fraction of Swedish LGBTQ+ books has an LGBTQ+ term assigned and therefore remain undiscoverable when searching using LGBTQ+ terms. Of the LGBTQ+ books with any tags assigned, only 5.2% (n = 25) had an LGBTQ+ term assigned in both genre and tags, 11.4% (n = 55) had an LGBTQ+ genre term, and 13.4% (n = 65) had an LGBTQ+ tag term.

Similar to Goodreads, all the 55 books with an LGBTQ+ term in genre use the same, broad “LGBTQ+” term, without further distinctions or more specific concepts. End users have earlier named this as a problem as their information needs are not being met when they want to distinguish between lesbian, gay, trans, etc. especially since male gay themes dominate LGBTQ+ fiction and other themes are less represented (Golub et al., 2022).

Further analysis shows that there is limited use of LGBTQ+ terms in social tags to denote relevant concepts, in contrast to the variety offered by the QLIT thesaurus. At the same time, inconsistency in tags is high as different synonyms are used in the same tag cloud, often at a broader level of specificity than appropriate for the work. Examples below are analyzed in detail to illustrate potential and challenges of LibraryThing tag clouds, in comparison to Queerlit indexing.

“Sen tar vi Berlin” by Moa-Lina Croall from 2006, in Queerlit at https://queerlit.dh.gu.se/work/vc55c0761ms8t30/sen-tar-vi-berlin, lists four QLIT terms:

1. “young lesbians” (unga lesbiska),

2. “mental disorders” (psykisk ohälsa),

3. “substance use (LGBTQ+)” (droganvändning (HBTQI)),

4. “love between women” (kärlek mellan kvinnor).

While the LibraryThing tag cloud (https://www.librarything.com/work/1701256) reflects these themes to some degree, it is characterised by inconsistency in tags and also requires post-coordination at the time of searching. There are three LibraryThing tags denoting LGBTQ+ identities:

1. “lesbian” which is broader than “young lesbians” assigned by Queerlit,

2. “homosexuality” which is even broader, and

3. “LGBTQ literature” (HBTQ litteratur) that is broader still.

Indexing inconsistencies place the burden on the searcher who would then need to think of all the possible terms denoting the same concept.

The tag cloud also lists the non-LGBTQ specific terms “young adults” (unga-vuxna) and “youth” – again applying both terms inconsistently to refer to the same concept. While this can be combined with “lesbian” at the time of searching, it does require the user to think of doing so. Another LibraryThing tag, “women”, could also be combined with “homosexuality” but would require the searcher to think about that combination, too. A further complication is that languages are mixed here – both English and Swedish are used in the tags, again implicitly placing the burden on the user at the time of searching.

Looking at the remaining two QLIT index terms “mental disorders” and “substance use (LGBTQ+)”, the LibraryThing tag cloud lists related terms “anxiety” and “alcohol”. The tag “anxiety” provides a more specific aspect of the QLIT concept “mental disorders” and could be considered a useful addition to the QLIT thesaurus as a narrower term or at least as a non-preferred term so that users who look for “anxiety” (ångest) are able to find a relevant broader term rather than getting zero results. “Alcohol” is accurate but the QLIT only has “alcoholism”, which would not be accurate for the book. Alternatively, the relevant LibraryThing tags could be added as non-controlled index terms. While “alcoholism (LGBTQ+)” (alkoholism (HBTQI)) exists in QLIT, the term is not linked to “substance use (LGBTQ+)”. Creating a link in the form of a thesaurus relationship could allow the end user searching for “alcoholism (LGBTQ+)” to expand their set of retrieved documents by choosing a broader or related term such as “substance use (LGBTQ+)”.

LibraryThing uses broader terms such as “love,” whereas Queerlit employs the more specific QLIT term “love between women.” Similarly, LibraryThing lists “identity,” while Queerlit uses the controlled SAO term “identity (psychology)” (identitet (psykologi)). Additionally, the term “sex” in LibraryThing does not conform to the 20% indexing rule.

Queerlit also assigns index terms representing non-LGBTQ+ related themes from the general Swedish subject headings system (SAO). For this work, these are: “Germany–Berlin” (Tyskland – Berlin), “travels” (resor), “musician” (musiker), women”s house (kvinnohus), “women political activists” (kvinnliga politiska aktivister) and “identity (psychology)” (identitet (psykologi)) as mentioned above. These concepts are partly represented in LibraryThing:

1. “Berlin”, which out of the country context may be ambiguous (e.g. it might refer to other places throughout the world, or people from arts, entertainment and media – see Wikipedia disambiguation page for “Berlin” (https://en.wikipedia.org/wiki/Berlin_(disambiguation));

2. “feminism” and “politics” which are broader terms related to the more specific and pre-coordinated term “women political activists”.

No terms denoting travels, musicians or women’s houses are available, so for users interested in these specific themes this work would not be retrieved in LibraryThing.

When it comes to genre, time and place, Queerlit assigns the following metadata: for genre, “novels” (romaner) as well as “Hc.01” which is a class taken from SAB (Swedish classification) and indicates “Swedish novels and short stories” – so, both genre and place facets (https://libris.kb.se/subjecttree_node.jsp?code=Hc.01). In comparison, LibraryThing has two different terms for the place facet: “Swedish author” (svensk författare) and “Sweden”, mixing both languages and choice of term, leading to inconsistencies and placing the burden on the searcher to think of the different terms. Genre term is “fiction” (skönlitteratur) which is broader than the concept used in Queerlit – “novels” (“romaner”). At the same time, two tags are added which cover concepts not covered in Queerlit: “debut book” (debutbok), “easy to read” (lättläst). The former might be a useful addition to users of Queerlit and SAOGF does not include this specific concept. The latter, however, seems to be a reflection of the tagger’s personal impression, as the book does not have characteristics of what is usually considered easy to read (short sentences, space between lines etc.).

Also, three tags referring to the same concept of time are added, providing conceptual duplicates: “2000s” (00-tal), 21st century, and “taking place during the 2000s” (utspelas 00-tal). This is also a duplicate with the metadata element of the publication year, 2006, which is why SAO indexing policies prescribe that time should only be indexed if the work is set in another time period than when it was published. Finally, there is one more tag not mentioned previously, “owns” (äger), likely indicating that the user has this book in their possession.

Similar findings characterise other LibraryThing metadata. “Bananflugornas herre” by Fredrik Eklund from 2005 (https://queerlit.dh.gu.se/work/r821qt4358p3k23/bananflugornas-herre) has LGBTQ+ themes from Queerlit partly reflected and inconsistently applied in LibraryThing (https://www.librarything.com/work/646642): Queerlit’s coordinated “porn films (LGBTQ+)” (porrfilmer (HBTQI)) and “porn industry” (porrindustri) have their corresponding non-coordinated tags in “porn”, “pornography” and Swedish “sexindustri” (sex industry being broader than porn industry, and thus a duplicate); as well as inconsistently applied “homosexuality” and “lgbt”. LibraryThing tag “sex” can be considered a duplicate or subsumed under “porn” or “pornography”. Queerlit”s “gay men” (bögar) is represented by “homosexuality” in LibraryThing. Queerlit’s terms “porn stars” (porrstjärnor) and “social norms” (sociala normer”) are missing in LibraryThing. The SAO term “personal development” (personlig utveckling) is not represented in LibraryThing, while “United States” (Förenta staterna) has a corresponding tag “USA”. While Queerlit uses the more specific term “love between men” (kärlek mellan män), LibraryThing uses a broad term “love”.

As for genre, place and time, Queerlit lists “novels” (romaner), “autobiographies” (självbiografiska skildringar), bildungsroman (utvecklingsromaner) and has two classifications, SAB: Hc.01 for Swedish novels and short stories as well as fiaf/2: F81 (FIAF Classification Scheme for Literature on Film and Television). LibraryThing has “Swedish writer”, “Sweden” and “Swedish” as conceptual duplicates. Genre terms are “biography”, “memoir”, as well as broader terms “fiction”, “novel” including its Swedish term “roman” (once in English and once in Swedish), as well as the Swedish term for fiction (skönlitteratur). The tag “specific people” (särskilda personer) likely refers to the genre of autobiography/biography as the book is about the author themselves. Indeed, Queerlit uses the SAO term “Fredrik Eklund”. LibraryThing also uses the term “easy to read” (lättläst) but, as in the previous example, it is more of a subjective interpretation by the tagger rather than what an easy to read book would be from a cataloguer”s perspective. The tag “bad language” (dåligt språk) is used by only one user account and it is not possible to understand its meaning without getting feedback from the account holder. There are two time facets which are different in meaning: 2000s (“00-tal) which might refer to the year of publication and as such is a duplicate; and “takes place in the 90s” (“utspelas 90-tal”). Like in the first example, there are two personal tags: “owns” (äger) and “read in 2009”. This is in spite of LibraryThing”s policy to remove personal tags (see the Background section).

The third example, “Du och jag, Marie Curie” by Annika Ruth Persson from 2003 (https://queerlit.dh.gu.se/work/j1trnglv16j8831/du-och-jag-marie-curie) has three main QLIT subject index terms: “coming out” (komma ut), “LGBTQ+ ex-partners” (före detta partners (HBTQI)), “mental disorders” (psykisk ohälsa); and three minor QLIT terms: “sex” (sex), “lesbian community” (lesbiska gemenskaper) and “parents of lesbians” (föräldrar till lesbiska). In comparison, the LibraryThing tag cloud (https://www.librarything.com/work/1554280) has only broader terms related to LGBTQ+: “lesbian”, “lesbian interest”, “lgbt”, “queer literature”, again presenting duplicates and inconsistencies while none of the six specific QLIT concepts are represented. Neither are any of the Swedish Children’s Subject Headings (books for children and young adults are not allowed to use SAO, but Swedish Children’s Subject Headings) represented: “LGBTQI” (HBTQI), “homosexuality” (homosexualitet), “coming out” (komma ut), “girls” (flickor), football (fotboll), “1980s” (1980-talet), “Marie Curie”, “Sweden–Gothenburg” (Sverige – Göteborg). It is important to note that since the Queerlit database is connected to Libris, the Swedish union catalogue that allows indexing by multiple libraries, subject headings from systems other than QLIT co-exist in the metadata record, even if they are conceptual duplicates.

Queerlit identifies this work as “books for youth” and “diaries” (ungdomsböcker, dagböcker) with SAB classification “Hc,u” representing Swedish fiction for young adults aged 13 and above. LibraryThing tags related to genre, place and time are: “fiction”, “novels etc.”, “fiction for children/youth” (skön barn/ungdom, whereby “skön” is short for skönliteratur). Conceptual duplicates include “teenagers”, “young adult”, “young adult literature”, “youth”; “2003” (year of publication duplicated from the other metadata element), “Swedish” and “Swedish literature”. uHc is the SAB shelf code for the “Hc,u” classification code.

Looking at LGBTQ+ tags only, the variety, like in Goodreads is small. Most common are the very general “lgbt” (29 occurrences) and “queer” (27) with a few instances of other general terms: “gay interest” (5), “lesbians” (5), “glbt” (3), “bisexuality” (2), “coming out” (1) and “transgender” (1).

In summary, some examples from the sample indicate that a certain number of tags might complement subject indexing of Queerlit and related information services. With more taggers, this number might increase. However, social tags are dominated by problems reported earlier in the literature. They seem to be inconsistently applied and can also be incomplete and inaccurate. In addition, they present a mixture of theme, time, place and genre facets as well as personal tags, which is likely to confuse users. Social tagging services should consider editing the tags by distinguishing between these different facets systematically. Personal tags are not used and could be something that LibraryThing might consider showing only to users. Furthermore, offering a controlled vocabulary from which to choose terms could be offered in order to help increase consistency in terms, word forms and encourage more specific themes’ tagging (see Golub et al., 2014). Finally providing automatic suggestions might help users find focus for tagging (Golub et al., 2014).

Recent research has shown that users looking for LGBTQ+ fiction often turn to social media, personal networks or search engines to identify titles to read; whereas they come to libraries to conduct known-item searches and get a hold of a book identified using other channels (Golub et al., 2022). Searching by subject is particularly problematic because only a limited set of LGBTQ+ books are indexed with relevant keywords and the keywords tend to be broad or overly general (Golub et al., 2022). To tackle these issues, the Queerlit project (2021–2023) established a specialised database of Swedish LGBTQ+ fiction (https://queerlit.dh.gu.se/), utilising detailed and precise subject indexing with terms from the Swedish QLIT thesaurus dedicated to LGBTQ+ topics.

However, it is likely that many readers of Swedish LGBTQ+ fiction will keep using familiar platforms such as social media and search engines like Google to discover new books. Additionally, the metadata available on these platforms could offer supplementary indexing terms for professionals to consider. Consequently, this study examined: 1) the extent to which books from the Queerlit database were recognized as LGBTQ+ works in commercial platforms, specifically Google Books and two social tagging services – LibraryThing and Goodreads; and 2) whether the metadata on these platforms could enhance the detailed and specific indexing in the Queerlit database.

Google Books offers millions of books digitised through partnerships with libraries and publishers and pulls metadata from a variety of sources, including publishers, libraries, metadata aggregators and online resources such as Wikipedia. The automated metadata processes have resulted in issues with misclassification and metadata errors, as evidenced by earlier research. Goodreads, now owned by Amazon, gathers metadata from several sources and relies on volunteer librarians for accuracy, although user-generated tags or “shelves” are not being verified. LibraryThing provides cataloguing services to both individuals and libraries, importing metadata from Amazon and libraries, while allowing users to add personal tags. It attempts some control over tagging through a process called aliasing to link synonyms, ensuring some level of consistency in its tagging system.

Methodology comprised downloading all works found in the Queerlit database as of June 2023 start (n = 1,320). Of the 1320 LGBTQ+ works in Queerlit, only a small portion was found in the three web services: 8.26% on Google Books (n = 109), 13.26% on Goodreads (n = 175), while about half on LibraryThing (55.3%, n = 730). This underrepresentation of Swedish LGBTQ+ works in the three international commercial information services makes them of limited value to the readers.

The limited inclusion of Swedish books on social media reduces their discoverability, making it harder for users to find or recommend them. Fewer Swedish books also limit opportunities for community engagement among Swedish-speaking users. Additionally, recommendation algorithms prioritize larger markets, perpetuating the cycle of underrepresentation for Swedish books.

The above is exacerbated by the fact that only a minority of Queerlit books found in the three services are categorised as LGBTQ+. In Google Books, only 9.17% (n = 10) had an LGBTQ+ keyword. In Goodreads, 9.14% (n = 16) had an LGBTQ+ keyword listed on the book page while an additional 17.14% (n = 30) had at least one mention of an LGBTQ+ keyword among the less frequent tags found by following the “More” tags link; counting in the latter group shows that 26.28% of Queerlit titles found in Goodreads have an LGBTQ+ term. In LibraryThing, 11.4% (n = 55) had an LGBTQ+ genre term, and 13.4% (n = 65) had an LGBTQ+ tag term. Only 5.2% (n = 25) had an LGBTQ+ term assigned in both genre and tags.

Further analysis showed that Google Books index terms are much less specific and exhaustive than Queerlit terms. Google Books uses highly general terms, most referring to identities, such as “gay men”, “lesbians”, “gays”, “lesbian teenagers” and “bisexuals”. Similar findings were found in Goodreads and LibraryThing. LibraryThing uses the very general “LGBTQ+” index term. End users have earlier listed this as a problem as their information needs are not being met when they want to distinguish between lesbian, gay, trans, etc. especially since male gay themes dominate LGBTQ+ fiction and other themes are less represented. Using a low level of indexing specificity and exhaustivity leads to high recall with low precision, and is likely to result in many search results which the user has no option of further narrowing down. This is particularly a problem in large web search engines and databases with many documents.

The study has also confirmed earlier research results pointing to problems of Google Books’ automatically gathered metadata from a variety of sources, leading to problems such as inconsistencies in genre terms. For example, for the genre metadata element this includes using specific terms like “mystery” on some works and general terms like “fiction” on others as well as different term forms, such as “fiction” and “fiction/general”. In addition, some books have genre terms entered under the subject metadata element rather than the genre metadata element, such as “Swedish fiction”, or “Children’s literature, Swedish”. Also, inconsistent indexing is found in Goodreads, and the established indexing guideline to assign the most specific index term is not followed. For example, index terms such as “Fiction” and “Literary Fiction” which are near synonyms can be found on the same metadata record. Other examples of conceptual duplicates are the terms “LGBTQ” and “Queer”, also found on the same metadata record. Another problem is mixing of subject and genre, as in the index term “Gay Fiction” on a metadata record that additionally has both “Gay” and “Fiction” already.

Detailed analysis of Goodreads metadata records also showed the established problems of social tagging: different terms are used to refer to the same concept (“lgbt”, “lgbtq”, “queer”, “glbt”, “lgbtqia”), there is no word-form control, and there are tags for personal use (“tbr-lgbtq”). This leads to challenging information retrieval that requires searchers to think of all possible synonyms and word forms in order to retrieve relevant results. At the same time, the terms suffer from the lack of specificity – usually only most general terms denoting the LGBTQ+ identities are used.

In spite of LibraryThing having more indexing policies and index term control than Google Books and Goodreads, there is a high degree of inconsistency in the tags, resulting in conceptual duplicates and both broader and narrower terms in the same tag cloud. Aliasing appears to be limited to spelling variations, which is an improvement over Goodreads. However, the presence of synonyms and conceptual duplicates continues to hinder effective information retrieval, creating challenges for end users. Synonyms are also present, requiring users to manually combine terms during searches, which adds complexity for the searcher. While some LGBTQ+ themes from Queerlit are partially represented, the LGBTQ+ terms are rather broad and unvaried. Another issue is the use of mixed languages in the tags, with both English and Swedish appearing. This further increases the burden on users during searches. Personal tags are also found in some of the studied examples, despite LibraryThing’s policy to remove personal tags.

Still, some examples from the studied social tagging services indicate that certain tags might complement subject indexing of Queerlit and related information services. With more taggers, this might increase. However, social tags are dominated by problems reported earlier in the literature. They seem to be inconsistently applied, and can also be incomplete and inaccurate. In addition, they present a mixture of theme, time, place and genre facets as well as personal tags, which is likely to confuse users.

The limited number of Swedish books on social media might have impacted the tagging by potentially reducing the diversity of tags, as there are fewer users contributing. Significantly larger group of taggers might produce more tags of potential use to enrich the Queerlit metadata records. At the same time, it does not seem that larger numbers of taggers would have greatly improved the issues related to tagging such as control of form, consistency, specificity etc.

Social tagging services should consider editing the tags by distinguishing between the different facets of theme, time, place and genre systematically. Personal tags are not used and could be something that LibraryThing could consider showing only to users. Furthermore, offering a controlled vocabulary from which to choose terms could be offered in order to help increase consistency in terms, word forms and encourage more specific themes’ tagging (see Golub et al., 2014). Finally providing automatic suggestions might help users find focus for tagging (Golub et al., 2014).

The Queerlit database might benefit from consulting social tagging services when indexing both LGBTQ+ specific and general themes; an approach which should be more systematically evaluated. In addition to social tagging, the value of complementing professional indexing with automated approaches, including Generative Pre-trained Transformers (GPTs), should be researched in the future. The professional, social and automated approaches might provide a more complete metadata records, as far as they are checked for quality and alignment with the indexing policy at hand. Finally, user search interfaces should be enhanced to leverage semi-automated word-sense disambiguation, related term search to increase recall and narrower term search to increase precision. To do this effectively, much more research is needed to determine the best approaches in interface design.

Main contributions of this study lie in its highlighting the challenges of indexing Swedish LGBTQ+ fiction on platforms like Google Books, Goodreads and LibraryThing, confirming earlier findings of underrepresentation, inconsistent metadata and overly broad tagging. It underscores how these issues complicate information retrieval for users, as metadata often lacks specificity and includes conceptual duplicates, such as “LGBTQ” and “Queer.” While social tagging has potential to complement professional indexing, problems like inconsistent application, synonym duplication and mixing of themes hinder its effectiveness.

The study also suggests refining tagging systems by introducing controlled vocabularies, tag suggestions and clearer distinctions between thematic, temporal and genre-related tags. User search interfaces could be enhanced with semi-automated tools, such as word-sense disambiguation and related term searches, to balance recall and precision in large databases. Additionally, the study highlights the need for multilingual and localized indexing strategies to address the challenges of mixed-language tags in databases like LibraryThing. It calls for greater community engagement to enhance the diversity and quality of social tags, particularly for underrepresented categories like Swedish LGBTQ+ fiction.

By providing a focused analysis of the Queerlit database and comparing it to commercial platforms, the study empirically validates earlier observations while offering new insights. It contributes to a potential framework for integrating professional, social and automated approaches to improve metadata completeness and accuracy. Moving forward, further research is needed to contextualise these recommendations and explore their practical application to benefit underrepresented literary categories as well as other types of information resources.

This research was partially funded by Riksbankens jubileumsfond (IN20-0013). Thanks are due to Siska Humlesjö and other subject indexers from the Queerlit database, who have helped with interpreting some of the results and their potential value for Queerlit.

Figure A1