Mallet: features contribution on each prediction

Question

I'm developing a NER system on Mallet using CRFs.

Do you know if it is possible to collect the features contribution for each prediction? I need to know and understand the precise behavior of the CRF model.

Any suggestions?

Thanks.

Cheers, ukrania

Answer 1

Yes you can do this, just do print of CRF model and check weights of model, if value is big by absolute value then it has big influence in classification result.