The project is not shipping models, and the NER models on the SourceForge
site don't work very well. I suggest you take a look at OntoNotes4, with
that you can train quite good models.

Jörn

On Sat, May 27, 2017 at 1:15 PM, Damiano Porta <[email protected]>
wrote:

> Jorn what corpus are you using to build the main english models?
>
> 2017-05-27 13:14 GMT+02:00 Joern Kottmann <[email protected]>:
>
> > I don't know, for that corpus you have to order the Reuters data, but we
> > have formats support for it, should be easy to measure when you have the
> > data.
> >
> > Jörn
> >
> > On Fri, May 26, 2017 at 6:16 PM, Damiano Porta <[email protected]>
> > wrote:
> >
> > > Jorn, what is the current performace with CONLL 2003?
> > >
> > > 2017-05-26 17:43 GMT+02:00 Joern Kottmann <[email protected]>:
> > >
> > > > Hello,
> > > >
> > > > can you post performance numbers? Only if it helps with some data set
> > it
> > > > would make sense to add it.
> > > >
> > > > Jörn
> > > >
> > > > On Thu, May 25, 2017 at 3:10 PM, Damiano Porta <
> [email protected]
> > >
> > > > wrote:
> > > >
> > > > > Hello,
> > > > > do you think a StemmerFeatureGenerator can be useful for NER
> models?
> > > > > I can create a PR for it.
> > > > >
> > > > > Damiano
> > > > >
> > > >
> > >
> >
>

Reply via email to