extractive summarization using textrank #139
Conversation
|
implementation for issue #138 |
|
Now that I look at the Gensim implementation of TextRank, I see that it also includes an optional Also, this is not the original TextRank algorithm, but a variant that uses BM25 (and a lot of custom text cleaning under the hood). Are we okay with that? |
|
@lmdehaas The reason I stuck with ratio, is that on a sentence level I find the word count parameter less transparent and less useful --> you don't know how long your sentences are so you don't know the exact number of words you are going to need. On the other hand if we have a use case where we want a fixed summary length, word_count makes more sense. Do we envision such a use case? Then I will add the word_count parameter. I personally have no problem with it being a variation on the textrank algorithm, as my goal was also to simply exploit the gensim implementation as we already wrapped other gensim functions, but if this is a problem, please let me know (we can also rename it 'GensimTextrank' for clarity. |
|
I can definitely imagine such a use case; it's the most common use case for summarization benchmarking, so I would prefer including that parameter. Concerning the alternative Textrank implementation: I don't mind, I just wanted to make sure we were aware of this! |
Textrank was implemented using the gensim framework in order to get a list of extracted sentences from a document that could serve as a summary of the document.