[Question]: How to programmatically get the NER of a word?

[AngledLuffa]

Question

In the sample NER, you suggest doing the following:

# make example sentence
# use_tokenizer means split on whitespace?
sentence = Sentence("George Washington went to Washington", use_tokenizer=False)

# predict NER tags
tagger.predict(sentence)

# print sentence
print(sentence)

Is there a good way to get the NER of a specific word, such as the first word, George? The closest I can figure out is to look at the spans and their tokens, which have an idx field. Still, it would be much easier to look at the members sentence, say sentence[0], and somehow get the NER off it... sentence[0].ner does not exist, though

Thanks!

[AngledLuffa]

This is a flair.data.Label:

sentence[0].get_label('ner')

but it doesn't actually have anything on it

This is also a Label:

entity = sentence.get_spans('ner')[0]
entity.get_label('ner')
label.value    # this is PER
entity.tokens[0].idx    # this is 1, representing George, index in the array + 1

I think the NER documentation would benefit from an example of how to access it programmatically, and if there's an easier way, it would be good to hear about it. Thanks!

[helpmefindaname]

Hi @AngledLuffa ,
we are currently restructuring our docs and try to set them up in a more complete way soon.
However in the specific case, you have the answer to your question documented here

[AngledLuffa]

Thanks! That is definitely helpful. Might also be worthwhile to mention how to get the start and end indices of the words which were labeled.

…

On Fri, Jun 9, 2023, 3:34 PM Benedikt Fuchs ***@***.***> wrote: Hi @AngledLuffa <https://github.com/AngledLuffa> , we are currently restructuring our docs and try to set them up in a more complete way soon. However in the specific case, you have the answer to your question documented here <https://flairnlp.github.io/docs/tutorial-basics/how-predictions-work#getting-the-predictions> — Reply to this email directly, view it on GitHub <#3261 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA2AYWNC4AZT2OMUF67UGZDXKOQF7ANCNFSM6AAAAAAZABWZMU> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[nikhilcms]

Hi,

Can you pls provide a solution how to get actual start and end indices such that using this indices I can extract detected entity like spacy ner model provides ?