Updating PubMed


I just watched an interesting webcast by David Gillikin, Chief of NLM’s Bibliographic Services, about the upcoming changes to the PubMed interface, followed by extensive Q&A. There was some confusion about how existing functionality would be mapped to the new interface, and understandable concern that the familiar interface would become dramatically less so. From an outsider’s perspective, the changes that were implemented looked reasonable, reducing the clutter of the existing design with some simplified controls and a more modern look and feel.

Not everything was changed, however. In particular, the advanced search features that allow users to construct fielded searches and  to set up limits (among other things), were to remain unchanged. While some functionality that was previously on a different page would now be accessible directly from the main screen, advanced search was still segregated from the search results. This is unfortunate because from a Cognitive Dimensions perspective it makes the interface more viscous (forcing the user to switch back and forth) at same time that it introduces hard mental operations. A better design might have packaged the advanced search controls into a panel that could be opened alongside the search results, allowing more iterative query formulation.

One thing that struck me as interesting was what was not talked about. In the normal search interface, the text box into which users enter terms is small, having space for only about 40 characters. While that may be good enough for typical Google searches that consist of 1-3 words, it seems inadequate for PubMed, where some of the terms may be longer than 40 characters! The reason for having a longer search box is that eliciting longer queries has been shown to improve search performance (see, for example, Belkin et al. 2002 and 2003). It would be interesting to know if this aspect of the design was discussed, whether there were any requests for a larger search box, whether keeping it small had implications for query processing time, etc.

Finally, there was a passing references to a recommendation-like interface that purportedly showed similar searches made by other people. Somebody from Duke typed the following into the Q&A chat window:

Duke: will other user’s searches appear on the side near recent activity?
Duke: we did not like it!

The presenter didn’t know whether that feature would be available, and the topic was not followed up.  Poking around the current PubMed interface, I was not able to elicit that feature. This sounds like a classic Social Search interface, and I would love to know what the problems with the existing feature are and why people didn’t like it.


  1. The idea that other people might see your searches on medical subject matter raises lots of privacy hackles, both for ordinary people looking for medical information and for researchers concerned about divulging their intellectual property. I think people often overreact–you can make this work without compromising anyone’s privacy–but it is a concern that you have to address head on in this domain.

  2. Agreed. Showing aggregated term suggestions probably doesn’t compromise a person’s privacy if that person is not identified by the system. I could see offering people an option to share (“publish”) their search results: Have the system ask the person if they would be willing to save their findings (anonymously) to help out others in similar situations. It might work for lay people by appealing to their sense of empathy, and also by coupling that request with clearly-labeled recommendations derived from others’ queries.

    The issue is trickier with researchers looking for unexpected combinations of terms or concepts.

  3. I’m all for fancy features, but first I’d like to see snippets! And for open source articles, full-text search.

    I like the term “viscous” — that’s how it feels manually shifting to abstract view and searching for snippets (and perhaps not finding them because of thesaurus expansion) then switching back.

  4. Mark Johnson says:

    The suggestion feature you’re talking about doesn’t show other individuals’ searches. It shows the most common search terms, aggregated from past logs, related to the terms in your query. So there is no privacy issue.

  5. @Mark, thanks for the clarification. As I say, I wasn’t able to experience that UI myself, and was reacting to what was said by the people on the call. I do wonder, however, what their concern was. Was it that the suggestions weren’t effective? Or some other reason?

    @Rob Snippets and full-text search are certainly useful. Although if my memory serves correctly, at least some studies failed to show a difference in effectiveness of searches when comparing full text with abstract searching. I am sure that once again, it’s one of those things where different subsets of useful documents are retrieved by different techniques, and a combination of the two might do better.

    Regarding viscosity: many of the cognitive dimensions names are well-chosen and readily evoke the abstractions they are intended to represent. It’s a vocabulary worth internalizing for those engaged in user interface design.

Comments are closed.