Would you be interested in adding more modern extractive summarization methods using things like BERT?

[nbertagnolli]

I'm working on replacing sumy in an existing project with a BERT based summarization model. Would you be interested in me making a PR which adds a BERTSummarizer class to this repository? Basically using this: https://arxiv.org/pdf/1906.04165.pdf. It would add a number of additional dependencies and wouldn't be compatible with python 2.7. Just thought I'd offer while I was working on it : ).

[miso-belica]

Hi, that would be great. Python 2.7 is dead so it's not a big problem for me. But if you send a PR please take a look at how dependencies are done for LSA or other summarizes and add them as optional. You can see it in https://github.com/miso-belica/sumy/blob/main/setup.py#L53-L58

[nbertagnolli]

Will do! I'll take a stab at this in the next week or two : ). Thanks for the great library : )

[dennlinger]

Not sure if this is still up-to-date, but introducing BERT-like methods (or for that matter, any other NN), is also a bit trickier for the computational load.
Many of these approaches only work somewhat efficiently when having access to a GPU, and one of the key advantages of sumy in my opinion is that you quickly can get some summaries for a text :)