EBSCO users span the globe, and we understand that not only are languages different, but so are the ways in which folks identify and interact with data. Each user has their own mental model comprised of their experiences, their culture, their language, and their needs.
This is why supporting equitable search has been a priority for EBSCO for decades, starting with the mapping of publisher subject headings across databases to ensure no matter which publisher subject vocabulary a user was familiar with, they could still retrieve content, even if the subject tags were not a synonymous match. This mapping is called the Unified Subject Index, or USI.
EBSCO took this a step farther and added all National Library Subject Authorities and the most authoritative government and linked data vocabularies to the USI, which created one of the largest multilingual mappings of scholarly vocabularies in the world, covering more than 280 languages and dialects. Even better, that same year we gathered billions of natural language terms to map to their controlled term equivalencies within the USI, and conducted bidirectional card sorting exercises to gather additional terminology from survey participants to map into the USI. Many of these vocabularies are updated and mapped in as linked data, and everything is stored as a knowledge graph for easier graph traversal (faster query time).
But, why so much mapping? Doesn’t a search engine automatically expand a query into synonyms, and now can’t AI just do all this itself?
Well, no, not exactly. Commercial search engines and AI both derive synonyms from common dictionaries, either from the commercial search engine itself, or the general intelligence found on the open web in the case of AI. Neither of which are likely to be equitable, with much of the open web favoring certain languages and certain perspectives, the lack of scholarly vocabularies in formats the commercial search engines and AI can easily ingest and understand, not to mention slurs found in both. AI cannot make the distinction between user warrant, literary warrant (the most common in library science during cataloging and indexing), and scientific warrant.
The USI mapping purposely covers all three. User warrant is covered by our users' natural language terminology, literary warrant is covered by the publication subject authorities, and scientific warrant (that of the terminology used by the researcher) is covered by the terminology found in the full text during search retrieval. This is why that even with AI being used in the new EBSCO Discovery Service (EDS) and EBSCOhost (eHost) search parsing, the underlying search logic that includes the USI is imperative to use, and makes our AI search mode more equitable than what would be possible with AI alone.
In addition to the USI, EBSCO also is using AI in search to help “decode” the search experience for those new to academic search. Complex advanced queries are still the standard for advanced research, but many struggle to find a foothold if they are unfamiliar with research and their library resources. The new Natural Language Search mode in EDS and eHost helps break down the barriers to entry by helping parse the query into more meaningful noun-phrase chunks that helps the EBSCO proprietary search engine retrieve not only relevant results, but contextually specific results as well, honoring the users intended query. This helps users who may not know how to formulate a complex query to still retrieve results that will help them along their research journey. This helps level the field and allows more folks to get going on their research, without specialized knowledge, making for a more equitable search experience.
Adding to this, EBSCO also has a widely diverse set of content ranging across experience levels, methodologies, languages, domains of study, and cultural and microcultural epistemologies from the widest collection of journal titles. By grounding our AI features off of this diverse set of content (grounding is not AI training), EBSCO continues to support our dedication to an equitable research experience.
