Entities with roles not validated correctly during test/cross validation

Contributing to Open Source Contributing Code
1

Thanks for introducing the feature called Entity Roles and Groups. it has been great addition!

I have noticed that Entity roles are extracted along with their BILOU tags from CRFEntityExtractor, which is causing an issue during test/cross validation phase. So all the role based entities have 0 accuracy metric scores always.

Hope this can be solved by removing BILOU tags and combining multiple tokens within the entity value.

here are my data files:

datasample
datasample1366×511 45.3 KB

DIETClassifier_report:

report
report301×770 6.44 KB

DIETClassifier_errors: error1

3 Likes
2

Exact same issue here ! I have colors entities with different roles and the extracted roles are all prefixed by U-, therefore breaking all my stories

3

:up:

2 Likes
4

:up:

5

Thanks for fixing this in 1.10.13

  • #6577: Remove BILOU tag prefix from role and group labels when creating entities.