Search Mailing List Archives


Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

[bioontology-support] [BioPortal] Feedback from Jessie Tenenbaum

Marcos Martinez Romero marcosmr at stanford.edu
Wed Aug 3 23:06:53 PDT 2016


Dear Jessie,

I think there's a bug in the recommender algorithm- NCIT is the top recommended ontology EVERY time, even when I put text from an abstract about plant model organisms. Seems unlikely that's really the best ontology?

The Recommender results are strongly based on the fraction of input data that is covered by existing ontology classes. The Recommender invokes the Annotator to obtain all annotations for the input text and assigns a score to each one of them. All these scores are then aggregated into the 'coverage score', which represents a 55% of the final score for each ontology. Note that the importance assigned to the coverage score can be customized in the 'advanced options' section.

The NCIT is generally very good at annotating biomedical text because it contains classes for many general concepts, such as "twenty", "year", "history", "replaced", "blue", "light", "factors", etc. That is why NCIT is often highly ranked when sending abstracts to the Recommender. When running the Recommender for one of your abstracts, you will probably see that the number of annotations provided by NCIT is very high.

However, not all words in an abstract have the same importance. Ideally, the Recommender should be able to identify the most relevant words and make a recommendation based on them. However, this feature is not currently available. As an alternative to get more accurate results, you could extract the most relevant keywords for those abstracts (either manually or using an external tool) and then use the "Keywords (separated by commas)" input type.

It would also be helpful if the Score column names had tooltips or links to tell what they each mean. Final vs. coverage vs. Detail vs. Specialization?

That's a good idea. Thanks. We are also working on a publication that will describe the details of the recommendation approach used.

Thanks for your feedback,

Marcos


On Aug 3, 2016, at 6:43 PM, support at bioontology.org<mailto:support at bioontology.org> wrote:


Name: Jessie Tenenbaum

Email: jessie.tenenbaum at duke.edu<mailto:jessie.tenenbaum at duke.edu>

Location: http://bioportal.bioontology.org/recommender

Feedback:

I think there's a bug in the recommender algorithm- NCIT is the top recommended ontology EVERY time, even when I put text from an abstract about plant model organisms. Seems unlikely that's really the best ontology?
It would also be helpful if the Score column names had tooltips or links to tell what they each mean. Final vs. coverage vs. Detail vs. Specialization?
Thanks!

_______________________________________________
bioontology-support mailing list
bioontology-support at lists.stanford.edu<mailto:bioontology-support at lists.stanford.edu>
https://mailman.stanford.edu/mailman/listinfo/bioontology-support

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.stanford.edu/pipermail/bioontology-support/attachments/20160804/01e6459b/attachment.html>


More information about the bioontology-support mailing list