eval question

[ttslr]

Thank you very much for your code！

I run the Eval.py on the “sampleData/test_predictions_41000.txt”

But the result does not seem to be very high

How is the result (F1: 91.21 )in the paper obtained?

Thank you very much!!
Best Regards!

[LopezGG]

O has high P/R (98) . The score they report was overall not subdivided by type ( eg ORG,MISC and LOC). you might want to try (weighted) averaging. I have not tried that as yet. Do try and let me know.

[Jong-Won]

@LopezGG
In my opinion, the NE overall F1 is calculated by:
F1 = 2P*R/(P+R)
the overall precision is calculated by:
#correct NEs in result / #total NEs in result
they have already not included tag "O" when overall F1