PaperScorer — how predictable is paper acceptance?
Sachin Konan, Jonathan Liu and Zhuang Liu (Princeton), NeurIPS 2026.
PaperScorer takes the anonymised PDF of a submission and returns an estimate of its chance of acceptance — a number plus one of three labels (Leans Reject, Leans Accept, Likely Accept). The model reads page images, so figures and layout count as much as the text does.
What makes it worth keeping is the question the authors ask: how predictable is acceptance in the first place. Every prediction comes with a page explaining how much to trust it, and the paper is public. Used that way it is a cheap sanity check on a draft — an early warning that a paper reads thin — not a verdict on whether to submit.
Two things to keep in mind. The PDF is uploaded to someone else's server, so send the anonymised version and nothing you are not ready to make public. And the site itself warns that it is processing a lot of papers right now, so results can lag.