ctakes-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dligach, Dmitriy" <ddlig...@luc.edu>
Subject Re: Temporal module dictionary
Date Thu, 11 May 2017 18:12:25 GMT
Hi Erin,

I would expect the relation annotator to catch contains(for 4 years, sinusitis) but clearly
it’s far from perfect.

I am surprised it catches contains(4 years ago, *city*) but this is mostly because *city*
is marked as an event (not sure why). 

Dima



> On May 11, 2017, at 12:00, Erin Gustafson <erin.gustafson@northwestern.edu> wrote:
> 
> Hey Dima,
> 
> Yes, that is one solution! Perhaps that's just the best way to go.
> 
> I'm seeing right now that the annotator is missing some relations that I would expect
it to catch (unless I'm misunderstanding something, which is possible). For example:
> 
> He reports sinusitis and rhinitis for 4 years, which started when he moved to *city*
4 years ago.
> -> misses sinusitis for 4 years, rhinitis for 4 years
> -> catches *city* 4 years ago
> 
> Would you expect those sort of expressions to be captured by the temporal module?
> 
> Thanks,
> Erin
> 
> 
> -----Original Message-----
> From: Dligach, Dmitriy [mailto:ddligach@luc.edu] 
> Sent: Thursday, May 11, 2017 11:50 AM
> To: dev@ctakes.apache.org
> Subject: Re: Temporal module dictionary
> 
> Hi Erin,
> 
> Is it an option to use all events and then just post-process the output of temporal relation
extraction to include the events you are interested in? The temporal module may break if you
exclude some events.
> 
> Dima
> 
> 
> 
>> On May 11, 2017, at 11:44, Erin Gustafson <erin.gustafson@northwestern.edu>
wrote:
>> 
>> Hi all,
>> 
>> I would like to use the temporal module to detect temporal relations involving events
specific to a phenotype. I've created a custom .bsv dictionary with a limited set of concepts
relevant to that phenotype, which I have used in the past as input to the dictionary look-up
algorithm. Now I'd like to try to use the same dictionary with the temporal module to limit
the extracted relations to those involving events of interest.
>> 
>> Is it possible to do this? I've plugged my dictionary in to FullTemporalExtractionPipeline,
but the detected events still include concepts that fall outside my dictionary.
>> 
>> Thanks,
>> Erin
> 

Mime
View raw message