Gault, Lora V., Mary Shultz, and Kathy J. Davies. “Variations in Medical Subject Headings (MeSH) mapping: from the natural language of patron terms to the controlled vocabulary of mapped lists.” Journal of the Medical Library Association 90.2 (2002): 173-180.
Introducing their work, the authors note that many users of MEDLINE begin their searches with terms that are not authoritative MeSH. This means that the results they receive depend on their MEDLINE interface’s mapping process. This study was performed to determine whether the tested interfaces (Internet Grateful Med, PubMed MeSH Browser, PubMed Index/Preview feature, OVID, OCLC FirstSearch Index option (with MeSH Heading Phrase), and OCLC FirstSearch Index (with MeSH Heading)) will return the same mapped results to users. To measure this, the researchers pulled and parsed research questions posed at a university health sciences center, working to establish the broadest natural language terms possible for the concepts. They determined equivalent MeSH headings for as many of the concepts as possible, discarding the others, and entered the natural language terms into searches on the six interfaces. Once the terms had been entered, the searchers counted and recorded what subject headings were returned. This done, they checked whether the various interfaces had mapped the terms to the correct subject headings and, if so, where the correct mapping appeared in the list of suggested subject headings (and how long that list was).
The results revealed that the interfaces did in fact map to terms differently, with varying success rates: Online Grateful Med had the correct heading 79% of the time, while other interfaces matched it on fewer than 50% of searches. While the authors concluded that some of the failures came from poor structure within MeSH, others seemed to spring from failures of the interfaces to make full use of MeSH. One definite design flaw noted was that the technique of alphabetic mapping used by several interfaces often missed important headings.
Overall, the researchers found that not all MEDLINE interfaces are equal, a fact that can lead to very different search results for users. The interfaces tested did have a variety of strengths, however, which they suggested could be combined to provide better mapping in future.
The study was well planned, carefully conducted, and coherently explained. The information gained is understandable and readily applicable to medical librarianship. I liked this article a good deal.