Many Eyes and literature summary

04 Oct

I’m not the first one to come up with this idea – Ntino posted about it before. However, I didn’t really understand before how powerful it could be. Using Many Eyes visualization capabilities I’ve created a quick browsable summary of abstracts related to a particular protein. I took all abstracts PubMed returned for a particular query (in this case it was “YadA Yersinia”; YadA is a prominent adhesin and important pathogenicity factor in Yersiniae) and uploaded them as text into Many Eyes. I chose “Word Tree” representation and searched for “yada”, which gave a nice graph of the most prominent phases related to this protein/gene name. Maybe it’s not a breakthrough, but compared to the classification/semantification provided by GoPubMed, such approach works much better for entities that aren’t well described in biological ontologies.

Given that the whole concept is pretty straightforward, it would be nice if one of alternative PubMed search engines provided a similar method of summarizing user’s query, don’t you think?

Reblog this post [with Zemanta]

Posted by on October 4, 2008 in Papers, Research, Visualization


Tags: , , , , , , ,

9 responses to “Many Eyes and literature summary

  1. Deepak

    October 4, 2008 at 17:32

    That’s very neat. I hear you. We could do a lot better with visualization/representation of semi-structured data. Perhaps the folks at pubmed, or gopubmed could do something like that using one of the available APIs. Shouldn’t be that hard

  2. Rafael Sidi

    October 5, 2008 at 00:42

    I second Deepak. This is nice and building these kind of representation should not be difficult using APIs.

  3. Dennis D. McDonald

    October 6, 2008 at 12:53

    You state “…such approach works much better for entities that aren’t well described in biological ontologies.” Wouldn’t source text built according to structured rules actually perform better, especially if the volume of text is high?

  4. Pawel Szczesny

    October 6, 2008 at 13:15

    Deepak, I of course agree – we didn’t yet explored most of the possible ways of visualizing data in biosciences. My feeling is that with a proper tool that helps to wrap one’s brain around the problem, we could advance science much faster than it is possible now.

    Dennis, if I understood you correctly, I agree, it would perform better – if we had structured source text. Much of scientific literature abstracts has zero metadata assigned, no to mention any structure of the text. In many cases (when ontologies for particular part of biology are well built) we can mine it and make it somehow semi-structured, but there are areas almost without semantic descriptors.

  5. Hilary

    October 7, 2008 at 14:54

    I really like this visualization, however, I wish I could play around with it a bit more… can you provide a link to the version on the Many Eyes site?

  6. Richard Karpinski

    October 7, 2008 at 20:15

    Nice. Put what you did on a wiki and let others turn it into a script which captures an arbitrary search (even other than Pub Med) and dumps it into Many Eyes along with a key word in such a way that this presentation is created. Other people can process search results into other kinds of presentations, too.

    The referenced articles themselves can also be put into other useful arrangements. I’d be interested in pushing an article into the Question, Answers, and arguments PRO and CON form, which takes more work than with Many Eyes since you have to identify the answers and arguments. Even worse, you often have to construct the question using imagination and a careful reading of the text. Still, having the question clarifies the answer amazingly for the reader.

  7. Pawel Szczesny

    October 21, 2008 at 09:27

    Hilary, sorry, I’ve just noticed your request. Unfortunately, I cannot provide you with a link, since the data were deleted from ManyEyes just after writing this post. It’s because I didn’t want to run into licensing issues using abstracts of articles from non-OA journals.

%d bloggers like this: