Metadata
Authors: Toine Bogers, Vivien Petras ∙ Year: 2017 ∙ DOI: 10.1515/dim-2017-0004 ∙ URL: https://linkinghub.elsevier.com/retrieve/pii/S2543925122000973
Abstract
Book search is far from a solved problem. Complex information needs often go beyond bibliographic facts and cover a combination of different aspects, such as specific genres or plot elements, engagement or novelty. Conventional book metadata may not be sufficient to address these kinds of information needs. In this paper, we present a large-scale empirical comparison of the effectiveness of book metadata elements for searching complex information needs. Using a test collection of over 2 million book records and over 330 real-world book search requests, we perform a highly controlled and in-depth analysis of topical metadata, comparing controlled vocabularies with social tags. Tags perform better overall in this setting, but controlled vocabulary terms provide complementary information, which will improve a search. We analyze potential underlying factors that contribute to search performance, such as the relevance aspect(s) mentioned in a request or the type of book. In addition, we investigate the possible causes of search failure. We conclude that neither tags nor controlled vocabularies are wholly suited to handling the complex information needs in book search, which means that different approaches to describe topical information in books are needed.
[TLDR] It is concluded that neither tags nor controlled vocabularies are wholly suited to handling the complex information needs in book search, which means that different approaches to describe topical information in books are needed.
Highlights
“For example, LibraryThing tags have been found to contain subjective, contextual, and personal descriptions,[^30][^50] whereas CVs such as the Library of Congress Subject Headings (LCSH) are required to be more abstract, objective, and impersonal. Whereas tags will cover whatever comes into a user’s mind about the document content, rule-based CVs may only represent specific aspects (such as the main topics of a document) [^33]” (Bogers and Petras, 2017, p. 19)
“Combining Tags and CVs may even find relevant documents for requests, which would fail for either individual metadata collection.” (Bogers and Petras, 2017, p. 32)
“Unique tags will better address these complex information needs than Tags.” (Bogers and Petras, 2017, p. 32)
Summary/Takeaways
- Controlled Vocabularies combined with unique tags surfaces more relevant searches.