The Easiest Way to do Coreference Resolution with spaCy with spaCy-Experimental

[@tariqkhan1518]

Thankyou it has been very helpful

[@scivanpoon]

Thx. I have a task to apply sentiments analysis (based on different tuned models, similar to FinBert, but sentiment scoring related various topics). On a long text, I want to apply coref on it and replace all non-primary mentions with the primary mentions, next split the text into sentences. and apply various sentiment scoring on each sentence if applicable (based on another topic classifier). Should I replace the all head with the first or longer head within a span cluster? and perform a NER? How to integrate workflow of NER and coref? Thank you!

[@scivanpoon]

How to increase the token size when i load the pretrained model into spacy? nlp = spacy.load("en_coreference_web_trf", max_length=##)?

[@jylecrennan]

Hey, is there a way to iterate through the doc.spans dict that it returns? I am tryin to retrieve the lists of the actual coreferences.

[@mighty_duck]

how do we replace the ambiguous coreference nouns so they can be treated as the same nouns when building a knowledge graph? ie replace the 'it' from the head cluster with 'dog' from 'coref_head_clusters_1': [dog, it]'

[@AlexF623]

Is there a way to find the location of the detected spans in the original input? Using a simple string find() might run into problems with duplicates...

[@python-programming]

Thanks for the comment. You should be able to access the .start_char and .end_char or .start and .end of each span. This will tell you where in the doc the data sits without having to find it.

[@dandradejose]

Is this Coreference Resolution english only?

[@gbennett1000]

Thank you! ❤

[@edythethompson7671]

Which version of python are you using?

[@shmouel4747]

Hi, when running nlp = spacy.load("en_coreference_web_trf") I get the error "Can't find factory for "experimental_coref"." How can I solve this issue?

[@python-programming]

It sounds like you did not pip install spacy experimental in that environment. Tey installing that first

[@shmouel4747]

@@python-programming now it get me the error "module 'srsly' has no attribute 'read_yaml' " when running import spacy

[@scivanpoon]

Does it suppport GPU?

[@ChristopherBrown-bj4zl]

What happened to my comment?

[@python-programming]

I only see this comment. I didn't remove any comments. Perhaps it was flagged by KZbin?

[@ChristopherBrown-bj4zl]

@@python-programming Odd. I guess my post might have been a little "botish"? All I did was ask, "how does this compare to the coreference project on the spacy site" and provided the link. As to not anger the KZbin Gods, I won't add the link again. The project is called "neuralcoref".

[@wdonno]

@@ChristopherBrown-bj4zl , a very quick search suggests that neuralcoref is from 2017 and is limited to English only and Spacy versions 2.x. A thread in the blog linked in the video description was from someone looking to upgrade to the new coref solution as testing found it more accurate in a healthcare data setting. Will try to link to the message in a following reply.