- From: Matthias Samwald <samwald@gmx.at>
- Date: Tue, 12 Feb 2008 22:53:18 +0100
- To: "Matt Williams" <matthew.williams@cancer.org.uk>, "Peter Ansell" <ansell.peter@gmail.com>
- Cc: <public-semweb-lifesci@w3.org>, <holger.stenzhorn@deri.org>, <p.roe@qut.edu.au>, <j.hogan@qut.edu.au>
> I'd agree that to capture all the publication might be hard, but to only > capture this bit (I suspect the conclusion) wouldn't you need to find the > conclusion, and ignore the rest? Using the abstract only might help, but > not enough....in any case, there are other bits (e.g. which type of > bananas they used) that you might well want to capture. Automatically capturing only a very selective part of a biomedical article (the main results/conclusions) might be even harder than capturing everything in an un-selective manner. This is why I am more interested in elegant hybrid systems, where NLP aids human annotators during the annotation process, instead of doing everything from start to finish automatically. Humans are much better at judging relevance, especially when they are annotating their own text/data during submission to a journal or database. With a good annotation system that makes use of NLP, existing ontologies and auto-suggest features, I would estimate that the main facts of most biomedical articles can be annotated in a very rich manner in less than 5-6 minutes -- not only mentioning entities that appear in the text, but actually formulating new facts and relations. If we compare that with the time authors need to spend on various other aspects of article submission, such as formatting the layout, checking citations etc., it seems realistic that authors would be able and willing to create such annotations... especially when we can demonstrate that it increases the usefulness and the impact of the article/database entry significantly. Cheers, Matthias Samwald
Received on Tuesday, 12 February 2008 21:53:49 UTC