Training the named entity recognition tool - correcting mistakes

[warrencowan]

Hi guys, am a bit of noob to the NER tool so just getting my head around things and need some pointers on training.

I get how you can add custom entities as a list to train with, but how do you make sure the alogirthm has learned correctly and correct where it has gotten it wrong.

For example, I wan to add a class called colour. So I get the adding a 2 custom list that says brown,red,white,blue and colour.colour,colour,colour etc

But what if I pass text strings into the tool like red faced, walter white, hand turned blue, red or dead, tom brown, captain scarlet etc etc How is the process going to be able to determine the difference between an actual colour and a name. 

I dont see any description for how the process is learning and not just text matching, nor any process for how to refine and and train out inconsistencies so it disambiguates better on subsequent runs.

Also once uts been trained, does the background librray in the tool update so it works without the list of custom entities and in other workflow instances.

Can someone enlighten me or set me straight?

thnx

w

[AlexanderHumfrey]

Would be great to get some information on this, seems to be very little documentation around this tool.