Before writing rules for any pattern, please gather at least 100 samples for target pattern
Place all these context patterns in single text file and run ANNIE over this.
Observe features of your required pattern from Annotation Type (Token, Sentence, NE outputs etc) feature set generated by ANNIE PRS
Divide Single large problem into mutiple probelms. Build rules in higherarchy manner so outputs from one step (Phase)
will be use as Input to another Step (Phase). That will make rule more readable, easy to debug and it will increase performace
by cutdowning the large Annotation Types (Token etc) by eliminating second level phases.
Test Rule over first on 100 samples, dont write rule as hard coded it should be more generic to work on new instances as well
Keep Initial Sample patterns as Test bed for future modifications impact over rul. Make Test samples as GoldenSet so you can test
over these smaples in future.
Once final commited rules shoul seriallize. so you can save some time while loaidng those in Application.
Try to use Jape Plus for initilizinf rules if possible.