Does OpenNLP use some kind of a cutoff (significance/confidence level), and that only the Named Entities with confidence/significance level grater than this cutoff level is returned (others discarded)?
The reason to ask this is that the FISE (of IKS) returnes a confidence level. Is it a FISE/IKS thing or is it coming from OpenNLP?
The OpenNLP name finder can calculate a confidence for an entity it detected. This calculation is done by NameFinderME.probs(Span). Have a look at this method. You can also retrieve the per token prob with the NameFinderME.probs() method, could be useful if you want the arithmetic mean instead.
Hope that helps,
I confirm that the confidence level of TextAnnotation instances in fise are based on the probabilities of the spans as returned by NameFinderME.
Sign up for the SourceForge newsletter:
You seem to have CSS turned off.
Please don't fill out this field.