Article Summary 4: Variations in Medical Subject Headings (MeSH) mapping: from the natural language of patron terms to the controlled vocabulary of mapped lists (Gault, Shultz, and Davies)

Gault et al. compared the methods and associated effectiveness of mapping from patron terms (natural language) to Medical Subject Headings (MeSH) across 6 MEDLINE interfaces. The results indicate that different interfaces (or more appropriately, different methods of MeSH mapping) often return different results following searches using patron terms.

The authors collected patron search terms from the Library of the Health Sciences, University of Illinois, then applied MeSH headings to each search term. The patron terms were then used in searches in 6 separate interfaces, allowing a comparison of mapping success across the interfaces. “Mapping success” is investigated by determining whether the patron term was successfully mapped to the MeSH term, where the appropriate term appeared within the list of possible MeSH terms, and the total number of MeSH term possibilities returned.

The authors give a brief summary of MeSH utility, explaining that while the use of these terms increases the specificity and accessibility of returns, patrons often are dependent on interface mapping to arrive at appropriate MeSH terms (e.g., the mapping of “heart attack” to “myocardial infarction”). If the patron term is not directly mapped to a MeSH term, many interfaces provide lists of possible terms for use.

The six MEDLINE interfaces studied were Internet Grateful Med (IGM), two PubMed options (MeSH Browser and Index/Preview function), OVID, and two OCLC FirstSearch interfaces (Index option using MeSH Heading Phrase and Index Option using MeSH Heading). IGM searched and mapped individual words in the patron term (using the Unified Medical Language System metathesaurus as well as MeSH terms ), so returns often mapped appropriately to one word but did not adequately translate to the term as a whole. The PubMed MeSH Browser provides a list of possible terms if an exact match to the patron term is not found. The PubMed Index/Preview feature returns terms that match alphabetically to the first (sometimes only) word of the patron term. OVID maps patron search terms to a list of possible MeSH terms using a “tree or thesaurus” to match meaning. The OCLC FirstSearch Index returns a list of terms that surround the patron term alphabetically. The OCLC FirstSearch Index option using MeSH heading phrase returned words and phrases that surround the patron term alphabetically, while the Index Option using MeSH Heading only returned single-word terms (located around the first letter of the patron term).

The results from patron term searches indicate a wide variety of search returns among the interfaces, with IGM (no longer in service) showing the highest rate of return. Looking at those searches that were mapped successfully, the authors examined the placement of the correct term within the listed results (with the understanding that a term returned higher on the list would more likely be found and subsequently used). Concept-based interfaces (IGM, PubMed MeSH Browser, and OVID) showed higher mapping success, but of the terms successfully mapped the alphabet-based interfaces were as successful in placing appropriate terms high on the return list. Variation in list length (or possible number of matches to patron term) led the authors to question if an optimum list length exists (ensuring appropriate return but not overwhelming the user).

Of the concept-based interfaces, differences in success rates are attributable to several possibilities. In some cases the MeSH term-linkage structure was not utilized by the interface (even if it existed in printed MeSH vocbularies), and in others the patron term did not have an adequate MeSH counterpart.

 This article highlights several things that librarians should keep in mind: identical searches of the same database will not yield identical returns if they are performed using different interfaces; knowing how the interfaces map to controlled vocabularies is an important aspect of success rate, and in many cases may alter how searches are performed; there is not necessarily a “best method” (although it can be said that concept-driven interfaces are more successful than alphabet-based approaches), and issues of precision and accuracy must be matched to patron needs when selecting an interface.

Gault LV, Shultz M, Davies KJ. Variations in Medical Subject Headings (MeSH) mapping: from the natural language of patron terms to the controlled vocabulary of mapped lists. J Med Libr Assoc. 2002; 90(2): 173-180.

 

Advertisements

7 thoughts on “Article Summary 4: Variations in Medical Subject Headings (MeSH) mapping: from the natural language of patron terms to the controlled vocabulary of mapped lists (Gault, Shultz, and Davies)

    • “Concept based interface” that’s interesting. I should look up the definition for that. What other kinds of interfaces are there. Being able to take people’s terms and then be able to find relevant articles is critical. I should do some research; I would like to know “the nuts and bolts” behind search engines. I know they have crawlers and algorithms, but still don’t really exactly understand what an algorithm is.

      • Well, Google (and other search engines) definitely has crawlers and several algorithms, but I’m not sure if the term crawlers applies to databases. From the way I understand it, crawlers will find/collect the information from a webpage, then a series of algorithms shuffles that information around to order the way results are displayed. On a very simple scale, if the search engine were responding to the search request “dog”, it would “crawl” around the internet, assess how many times “dog” appeared on different webpages, then use an algorithm to order the results display (so that the one with the most “dog”s would appear first, followed by the one with the next largest number, etc. In order to make the search more productive for the user, there would be MANY of those types of comparisons among webpages (so not only # of times “dog” appears, but also the date of last update, number of times other webpages link to that page, inclusion of “dog” in the metadata, etc.). Ultimately all of those algorithms (that are ordering returns) are combined to list the most relevant links first. In my mind this is clear (my description, that is – not algorithms in general), but I think that may all sound like garbage.
        I think “crawling” is the action of going to one page, then using the information on that page to go to other pages (so that all of the hyperlinks from one page will be visited, maybe metadata elements will be used to search other pages, etc.)
        Crystal clear, right? 🙂
        I believe there is a Google video that describes the basics of their search process, I’ll see if I can find it.

  1. Evelyn I somehow just saw your comment above…I’m not sure if you are still looking at the LS534 blogs or not, but to answer your question about how databases know if something is relevant/know how to order returns… I’m not sure they DO know. It is up to the searcher to structure the search to get meaningful returns. So we can use subject headings/controlled vocabulary to get exactly the topic we want (e.g., Mercury-metal instead of Mercury-deity). In the Mercury example, if mercury is used in a keyword search, the database would return anything that had Mercury of any meaning associated with the article. If we switch to using controlled vocabularies and subject searches we can essentially tell the database how to narrow the results. Then further restrict what the database would find relevant by including date ranges, source type, etc.
    As far as how the database ranks the returns (that are restricted to any level), I have no idea. Maybe it is different for different databases – like one returns them in chronological order but another one alphabetically by author or title? I know it is frustrating to do a search and have no idea why the results are listed as they are – like putting in an exact phrase you are searching, then not seeing that result in titles/abstracts until pages in to the result. I don’t get it.

    Sorry if this has been unhelpful, or if it comes across as condescending in any way. I can’t tell exactly what you are asking, and I clearly have some confusions of my own 🙂

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s