What was this blog post on recent developments in text summarizers?

October 28, 2019 5:35 PM

Within the past couple weeks I read something (a blog post, I think) describing how recent sophisticated, deep learning based text analysis methods did essentially no better than very simple algorithms. The point that stuck out was that for text summarizers, it was very hard to beat a summary consisting of the first three sentences of the document. Does this ring a bell?
posted by molla to Computers & Internet (4 answers total)

It's hard to say, but it might be this student presentation (PDF) which seems to make the same point. Of note, the conclusion says, "Significantly outperformed the abstractive state-of-the-art result."
posted by Meldanthral at 6:29 PM on October 28, 2019

Not ringing a bell, but starting with https://blog.fastforwardlabs.com/2016/04/11/new-tools-to-summarize-text.html and working forward might get you to something similar.
posted by dws at 8:55 PM on October 28, 2019

Mod note: One deleted. Sorry, but OP is looking for the actual blog post rather than an explanation of why this might be.
posted by taz (staff) at 2:26 AM on October 29, 2019

Response by poster: Found it! Thanks for the links, they look interesting too.
posted by molla at 7:52 AM on October 29, 2019