Hi Jairo,
which version of opennlp did you use?
I am asking since I had similarly poor performance on a german sentence
training scenario with opennlp-incubation-1.5.2.
The problem was that this version did only recognize "!", "." and "?" as
end of sentence symbols.
The current svn trunk has a version where you can specify your end of
sentence (EOS) character. Using this, the performance of my sentence
tagger also improved from around 0.7 with 1.5.2 to about 0.97 (with the
svn trunk).
Best
Katrin
On 02/13/2012 06:13 PM, Jairo Sarabia wrote:
Hello,
Responding to questions from Jörn:
- What training data do I use?
I've created my own data file:
I wrote every spanish sentence in a different line as the Spanish grammar. For
example, I created a new sentence after . or after , or after ? or after ! or
after : characters...
- Which settings did I use?
I've followed the steps of
https://sourceforge.net/apps/mediawiki/opennlp/index.php?title=Sentence_Detector
Specifically I ran:
*bin/opennlp SentenceDetectorTrainer -encoding UTF-8 -lang es -data
~/Documentos/es-sent.train -model es-Sent.bin -iterations 1000*
for training, and
*bin/opennlp SentenceDetectorEvaluator -encoding UTF-8 -model es-Sent.bin
-data ~/Documentos/es-sent.train*
for evaluation.
Grateful that you could help me and tell me what I can improve.
Cheers,
Jairo
2012/2/11 Jörn Kottmann<[email protected]>
These numbers indicate that it doesn't work at all.
Can you tell us a bit more?
- What training data do you use?
- Which settings did you use?
Jörn
On 02/10/2012 07:02 PM, Jairo Sarabia wrote:
Hello all!,
I've trained a spanish sentence detector with Training Tool
SentenceTrainerDetector function.
But when I evaluate it with EvaluatorTool and SentenceDetectorEvaluator
only returns 0.5 of precision and recall.
How can I improve precision and evaluation in general?
Thanks in advance!,
Jairo Sarabia
--
Dr. Katrin Tomanek
Averbis GmbH
Tennenbacher Strasse 11
D-79106 Freiburg
Fon: +49 (0) 761 - 203 97696
Fax: +49 (0) 761 - 203 97694
E-Mail: [email protected]
Geschäftsführer: Dr. med. Philipp Daumke, Dr. Kornél Markó
Sitz der Gesellschaft: Freiburg i. Br.
AG Freiburg i. Br., HRB 701080
