ctakes-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From samir chabou <samir...@yahoo.com>
Subject Re: implementation of << affects>> relation
Date Thu, 10 Oct 2013 19:49:42 GMT
Thanks Dmitriy for the article. I read it attentively and found it veryinteresting. 
 
In the conclusion of the article you have mentioned that your system is easily extendable
to other relationssuch as manages/treats (which is a sub relation of affects).
 
As part of my learning curve I’ll be interested in doing the extension and retrain your
classifier to include manages/treats under your lead and coaching. This task will be beneficial
for our team to gain practical experienceon how svm works,and at the same timeit shouldcontribute
to enhance the quality of the alignment relationbetween the medical question and the PICO
structure. Also, this experience will certainly boost my participation inthe cTAKES community.
 
On the base of your paper I suggest the following approachfor addingthe manages/treats relation:
 
·        Training corpus: SHARP and ShARe (how couldI get a copy of them, if possible)
 
·        Set of features related to the entities ofthe manages/treats relation: 
thesame features that you are using for the LocationOf and DegreeOf relations
 
·        Classifier parameters: need to be identified (probablythe same parameters
asyours;  I don’t know yet how these parameters can be set)
 
·        To retrain your (binary linear) classifier for manages/treats relation: I
need to know the tool used for this. Is it MALLET that you have used or some thing else?
 
·        Test corpus: sub set of the SHARP and ShARe in order to be able to quantify
the results (f 
score, precision and recall) since these corporaare already annotated
 
·        To validate the results
 
·        If validated, to deploy the manages/treats relation in ctakes-relation-extractorReply,




On Tuesday, October 8, 2013 9:31:28 AM, Dmitriy Dligach <dmitriy.dligach@childrens.harvard.edu>
wrote:
 
Hi Samir,

Before we talk any further, I'd like to mention that the relation 
extractor is a machine learning system. You can read more about it here:

http://jamia.bmj.com/content/early/2013/10/03/amiajnl-2013-001766.full

So if you want to extend it, you'll need some data that annotates the 
"affects" relation. Do you have this kind of data?

Dima


On 10/07/2013 05:38 PM, samir chabou wrote:
> Hi Pei, Dimitry and any other that may help on
> this …
>
 Currently I’m using cTAKES to semantically   analyze
> a medical question and then try to map it to PICO structure based on different
> medical type (relations, NE) identified in the question. I’m interested in extending
> the ctakes-relation-extractor project in order to handle the << affects>>
relation of the UMLS
> semantic network.
> First of all is that possible?
> If yes,
> what is the involvement?  My plan is:
> 1)      To proceed with the implementation locally and
> then check it in to be validated and deployed later (if it’s approved).
> 2)      How ever I’m not very familiar with the subject
> matter on how UMLS semantic relations are implemented in cTAKES. Is the <<
 affects>>
> relation will follow the same logic than LocationOfRelation  and DegreeOfLocation? 
Which kind of IdentifiedAnnotation that the <<
> affects>> relation will relates?  It
> would be very appreciated if some one guides me on this matter.
>  
> Thanks Samir

-- 
Dmitriy Dligach, PhD
Research Fellow
Children's Hospital Informatics Program
Boston Children's Hospital and Harvard Medical School
(617) 919-3596
Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message