Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
The ixa-pipe-nec [1] provides good quality Named Entity Recognition models for English, Spanish, Dutch, German and Italian. However to use those models one needs
- OpenNLP 1.6.0
- OpenNLP extensions provided by the ixa-pipe-nec module.
OpenNLP 1.6.0 is not yet released so we will need to go with a SNAPSHOT version for now. The ixa-pipe-nec module does not support OSGI. So we will need to embed the required classes into a bundle and provide a bundle activator that registers the extensions as OSGI services (with the metadata expected by OpenNLP).
NOTE: This issue will only cover extensions to Apache Stanbol so that one cane use the provided models. To use the models Users will need to download the ~700Mbyte archive linked on [1] get the OpenNLP modles (*.bin files) and put them into datafiles folder of Apache Stanbol.
The models use PER, ORG, LOC and MISC as types. So using a configuration for the CustomNERModelEnhancementEngine should do the trick:
# Configuration of org.apache.stanbol.enhancer.engines.opennlp.impl.CustomNERModelEnhancementEngine-ixa_nec.config stanbol.engines.opennlp-ner.typeMappings=["PER\ >\ http://dbpedia.org/ontology/Person","ORG\ >\ http://dbpedia.org/ontology/Organisation","LOC\ >\ http://dbpedia.org/ontology/Place","MISC\ >\ skos:Concept"] stanbol.enhancer.engine.name="ixa-nerc" stanbol.engines.opennlp-ner.nameFinderModels=["de-clusters-dictlbj-conll03.bin","en-91-18-4-class-muc7-conll03-ontonotes-4.0.bin","es-clusters-dictlbj-conll02.bin","it-clusters-evalita09.bin","nl-clusters-dictlbj-conll02.bin","eu-clusters-egunkaria.bin"]
The names of the OpenNLP model files are the values of the stanbol.engines.opennlp-ner.nameFinderModels property. You will find those files in the NERC-Models 1.5.0 file. See the documentation on [1] for more details and other options.
Attachments
Issue Links
- depends upon
-
STANBOL-1423 Update to OpenNLP 1.6.0
-
- Resolved
-