Re: NER Features

[Damiano Porta]

Thank you Daniel! Hmm ok... I will try to guess what features are the most
important :)

2018-01-30 18:19 GMT+01:00 Dan Russ :

> Damiano,
>You are treading in some dangerous waters.  You need to open up the
> black box of the model and peek inside.  I open up the NER model and get
> the maxent model inside.  I make the context using the feature generator,
> then I get the context map inside the maxent model to lookup the weights.
> I am grossly oversimplifying the process because otherwise I would be
> writing a treatise on OpenNLP.  I would suggest looking at the find()
> method and align what that method does with my comments on the steps you
> need to take.
>
> Hope it helps...
> Daniel
>
> > On Jan 30, 2018, at 12:10 PM, Damiano Porta 
> wrote:
> >
> > Hello everybody,
> >
> > how can we understand what are the most important features during the NER
> > process? I mean.. when the TokenNameFinder selects a label is it possible
> > to retrieve the most important features too ?
> >
> > Thanks
> > Damiano
>
>

Re: NER Features

[Dan Russ]

Damiano,
   You are treading in some dangerous waters.  You need to open up the black 
box of the model and peek inside.  I open up the NER model and get the maxent 
model inside.  I make the context using the feature generator, then I get the 
context map inside the maxent model to lookup the weights.  I am grossly 
oversimplifying the process because otherwise I would be writing a treatise on 
OpenNLP.  I would suggest looking at the find() method and align what that 
method does with my comments on the steps you need to take.

Hope it helps...
Daniel

> On Jan 30, 2018, at 12:10 PM, Damiano Porta  wrote:
> 
> Hello everybody,
> 
> how can we understand what are the most important features during the NER
> process? I mean.. when the TokenNameFinder selects a label is it possible
> to retrieve the most important features too ?
> 
> Thanks
> Damiano

NER Features

[Damiano Porta]

Hello everybody,

how can we understand what are the most important features during the NER
process? I mean.. when the TokenNameFinder selects a label is it possible
to retrieve the most important features too ?

Thanks
Damiano