Honest question: how do people in NLP deal with the enormous stream of papers 📄 coming in on a daily basis? I would need three times the hours ⏳ to finish my reading list before new papers come in again 📚
Comments
Log in with your Bluesky account to leave a comment
Honest answer: I don't read anything from the enormous stream of papers unless they are *critical* to my current work or I am deeply interested in the topic.
It's also much easier to be recognized for your work if you focus on an underdeveloped topic and lay the foundations for future research. Then you're not one out of many, but the person who did xyz.
(Also nobody except PhD students has time for labor intensive work. So that's what I did back then.)
Don't look at it as a stream. I stick to reading from venues I care about (so that's 6-7 conferences and several workshops) so I have about 30-40 papers every other month or so.
It does take me a day or two to filter down to those 30-40, but I think the time saved is worth.
Sort of? If I know the problem well, or the methods in the paper are iterative rather than a big leap, I can skim over most of it and average to 1h-ish per paper.
The other trick is to get a grad student XD
Like, I'll read papers my advisor might like and filter even further down to 10-15 at most
I feel it gets easier with experience. Most papers are incremental, so (1) if you miss one, you catch up with following ones in the same line of work. (2) Many papers can be "read" in a few minutes. Skim the abstract and a few other bits, and from experience you know/guess what the rest is about.
i don't keep up with it and i don't read nearly as much as i want to :( i try to do project-specific literature searches and i'm now better at triaging and reading just abstracts or just skimming, but idk how relevant my broader strategies are beyond the phd/postdoc level
for completeness, my broader strategies are to be (mostly) aware of things my collaborators have done and are working on, and to get recs from them + recs surfaced in other ways like scholar alerts / bluesky feeds / looking at the proceedings of confs i attend and workshops i care about
I use an automated filter and then output structured markdown files to feed into my Obsidian knowledge graph. When I need to brush up on a topic I just go to the cluster and the linked content
Carefully scoping your interests and refining a sense of what a "good" paper looks like a few paragraphs in, so that you can differentiate what papers you're skimming for breadth vs. reading for depth
Comments
Not all topics are popular and fast moving! You can choose to work in an area that isn't as crowded, and carve out a space for yourself to explore.
(Also nobody except PhD students has time for labor intensive work. So that's what I did back then.)
It does take me a day or two to filter down to those 30-40, but I think the time saved is worth.
The other trick is to get a grad student XD
Like, I'll read papers my advisor might like and filter even further down to 10-15 at most