You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@ctakes.apache.org by Chase Master <ch...@gmail.com> on 2013/08/15 17:18:29 UTC
Assertion module
Hi,
I’m using v3.0-incubating (as a Maven dependency) with the default clinical
pipeline, which calls the Assertion module. I noticed that for something
like “Anna’s father has had knee pain for years”, the subject attribute
gets set to family_member. Is it possible to be more specific? Also,
does it assume that the subject of every statement is the patient unless
the phrase suggests otherwise? If the possible values are specified by the
prefix ATTR_SUBJECT_ in edu.mayo.bmi.uima.core.type.constants.CONST, is it
possible to meaningfully change them?
Also, the value of “confidence” is set by the assertion module. Is the
score referencing the assignment of subject, or whether it’s generic, or
the polarity, or other attributes?
Thanks
Re: Assertion module
Posted by Chase Master <ch...@gmail.com>.
Thanks for your responses! Yes, I meant modifying ATTR_SUBJECT to
something more specific like the example you gave.
On Thu, Aug 15, 2013 at 10:51 AM, Chen, Pei
<Pe...@childrens.harvard.edu>wrote:
> > ATTR_SUBJECT_ in edu.mayo.bmi.uima.core.type.constants.CONST, is it
> > possible to meaningfully change them?
> Could you clarify a bit more about meaningfully changing them? i.e. did
> you mean to write another classifier to make the subject attribute more
> specific? Such as "sibling, biological father/mother?"
> I presume you mean ATTR_SUBJECT* in:
>
> http://svn.apache.org/repos/asf/ctakes/trunk/ctakes-type-system/src/main/java/org/apache/ctakes/typesystem/type/constants/CONST.java
>
> You may find the below links helpful:
> ML:
> http://svn.apache.org/repos/asf/ctakes/trunk/ctakes-assertion/src/main/java/org/apache/ctakes/assertion/medfacts/AssertionAnalysisEngine.java
> And additional Rules:
> http://svn.apache.org/repos/asf/ctakes/trunk/ctakes-assertion/src/main/java/org/apache/ctakes/assertion/attributes/SubjectAttributeAnalysisEngine.java
>
> > -----Original Message-----
> > From: Wu, Stephen T., Ph.D. [mailto:Wu.Stephen@mayo.edu]
> > Sent: Thursday, August 15, 2013 11:49 AM
> > To: dev@ctakes.apache.org
> > Subject: Re: Assertion module
> >
> > Hi chase,
> >
> > >pipeline, which calls the Assertion module. I noticed that for
> > >something like ³Anna¹s father has had knee pain for years², the subject
> > >attribute gets set to family_member. Is it possible to be more
> specific?
> > See below.
> >
> > >
> > >does it assume that the subject of every statement is the patient
> > >unless the phrase suggests otherwise?
> > Yes.
> >
> > >If the possible values are specified by the prefix ATTR_SUBJECT_ in
> > >edu.mayo.bmi.uima.core.type.constants.CONST, is it possible to
> > >meaningfully change them?
> > CONST does indeed include all the possible values. It is possible to
> rewrite
> > the Subject annotator in the assertion module such that other values can
> be
> > assigned. However, you would be working without "meaningful" evaluation
> > on its accuracy. In other words, if you assign "family_member" we have
> > human-annotated cases of what that looks like, but if we come up with a
> > new method to assign "father" we have no way to figure out whether we're
> > doing a good job or not.
> >
> > >Also, the value of ³confidence² is set by the assertion module. Is the
> > >score referencing the assignment of subject, or whether it¹s generic,
> > >or the polarity, or other attributes?
> > I believe the "confidence" number was intended to be for the named entity
> > recognition (ctakes-dictionary-lookup). The newer version of assertion
> (in
> > trunk but not the default in ctakes-clinical-pipeline) does not assign
> this
> > value. Matt Coarr -- any comments on what the medfacts version does?
> >
> > hth
> >
> > stephen
>
>
RE: Assertion module
Posted by "Chen, Pei" <Pe...@childrens.harvard.edu>.
> ATTR_SUBJECT_ in edu.mayo.bmi.uima.core.type.constants.CONST, is it
> possible to meaningfully change them?
Could you clarify a bit more about meaningfully changing them? i.e. did you mean to write another classifier to make the subject attribute more specific? Such as "sibling, biological father/mother?"
I presume you mean ATTR_SUBJECT* in:
http://svn.apache.org/repos/asf/ctakes/trunk/ctakes-type-system/src/main/java/org/apache/ctakes/typesystem/type/constants/CONST.java
You may find the below links helpful:
ML: http://svn.apache.org/repos/asf/ctakes/trunk/ctakes-assertion/src/main/java/org/apache/ctakes/assertion/medfacts/AssertionAnalysisEngine.java
And additional Rules: http://svn.apache.org/repos/asf/ctakes/trunk/ctakes-assertion/src/main/java/org/apache/ctakes/assertion/attributes/SubjectAttributeAnalysisEngine.java
> -----Original Message-----
> From: Wu, Stephen T., Ph.D. [mailto:Wu.Stephen@mayo.edu]
> Sent: Thursday, August 15, 2013 11:49 AM
> To: dev@ctakes.apache.org
> Subject: Re: Assertion module
>
> Hi chase,
>
> >pipeline, which calls the Assertion module. I noticed that for
> >something like ³Anna¹s father has had knee pain for years², the subject
> >attribute gets set to family_member. Is it possible to be more specific?
> See below.
>
> >
> >does it assume that the subject of every statement is the patient
> >unless the phrase suggests otherwise?
> Yes.
>
> >If the possible values are specified by the prefix ATTR_SUBJECT_ in
> >edu.mayo.bmi.uima.core.type.constants.CONST, is it possible to
> >meaningfully change them?
> CONST does indeed include all the possible values. It is possible to rewrite
> the Subject annotator in the assertion module such that other values can be
> assigned. However, you would be working without "meaningful" evaluation
> on its accuracy. In other words, if you assign "family_member" we have
> human-annotated cases of what that looks like, but if we come up with a
> new method to assign "father" we have no way to figure out whether we're
> doing a good job or not.
>
> >Also, the value of ³confidence² is set by the assertion module. Is the
> >score referencing the assignment of subject, or whether it¹s generic,
> >or the polarity, or other attributes?
> I believe the "confidence" number was intended to be for the named entity
> recognition (ctakes-dictionary-lookup). The newer version of assertion (in
> trunk but not the default in ctakes-clinical-pipeline) does not assign this
> value. Matt Coarr -- any comments on what the medfacts version does?
>
> hth
>
> stephen
Re: Assertion module
Posted by "Wu, Stephen T., Ph.D." <Wu...@mayo.edu>.
Hi chase,
>pipeline, which calls the Assertion module. I noticed that for something
>like ³Anna¹s father has had knee pain for years², the subject attribute
>gets set to family_member. Is it possible to be more specific?
See below.
>
>does it assume that the subject of every statement is the patient unless
>the phrase suggests otherwise?
Yes.
>If the possible values are specified by the
>prefix ATTR_SUBJECT_ in edu.mayo.bmi.uima.core.type.constants.CONST, is it
>possible to meaningfully change them?
CONST does indeed include all the possible values. It is possible to
rewrite the Subject annotator in the assertion module such that other
values can be assigned. However, you would be working without
"meaningful" evaluation on its accuracy. In other words, if you assign
"family_member" we have human-annotated cases of what that looks like, but
if we come up with a new method to assign "father" we have no way to
figure out whether we're doing a good job or not.
>Also, the value of ³confidence² is set by the assertion module. Is the
>score referencing the assignment of subject, or whether it¹s generic, or
>the polarity, or other attributes?
I believe the "confidence" number was intended to be for the named entity
recognition (ctakes-dictionary-lookup). The newer version of assertion (in
trunk but not the default in ctakes-clinical-pipeline) does not assign
this value. Matt Coarr -- any comments on what the medfacts version does?
hth
stephen