I need to calculate the confidence score for predictions of a NER- BERT model. I know how to get the probability of the predictions for each token, but I need a score for a complete span after aggregation if the named entity has more than one token (I mean there are also some tokens with I-tags). I would like to know what is a good way to calculate such scores? is it ok to multiply the confidence scores of the first and last tokens in the span?
Many thanks for your help