CFE question

[Jörn Kottmann <ko...@gmail.com>]

Hi,

after Michaels post yesterday I stumbled over a CFE paper:
http://domino.research.ibm.com/comm/research_projects.nsf/pages/medicalinformatics.pubs.html/$FILE/CFE_sominsky-A4.pdf

Is it possible to write the generated features back into the CAS?

The reason I ask is, because it could be very interesting for NLP
engines like OpenNLP to pass features from the CAS through to
the underlying classifier.  This way it is very easy to experiment with
new features.

Jörn

Re: CFE question

[Igor Sominsky <so...@gmail.com>]

Jörn,

Yes, the features can be written back into CAS. Actually, the feature extraction is a 2 step process. On the first step, the features are extracted from CAS according to FESL spec (the feature values are strings in the format that is also specified by FESL) by an annotator engine and stored in annotations of a specific type. On step 2 a task specific annotation engine or CAS consumer accesses the pre-extracted features and processes them according to the task requirements. In our applications we used different types of such consumers/annotators, one of which output feature values to build models for machine learning and another used those models to apply extracted features against these models for classification. I guess such an approach fits your purposes if I understood your question correctly. 

Let me know if further questions

Regards, 
Igor

On Jun 2, 2011, at 5:42 AM, Jörn Kottmann <ko...@gmail.com> wrote:

> Hi,
> 
> after Michaels post yesterday I stumbled over a CFE paper:
> http://domino.research.ibm.com/comm/research_projects.nsf/pages/medicalinformatics.pubs.html/$FILE/CFE_sominsky-A4.pdf
> 
> Is it possible to write the generated features back into the CAS?
> 
> The reason I ask is, because it could be very interesting for NLP
> engines like OpenNLP to pass features from the CAS through to
> the underlying classifier.  This way it is very easy to experiment with
> new features.
> 
> Jörn