This idea is part of the A Dollar Worth of Ideas series, with potential open source, research or data science projects or contributions for people to pursue. I would be interested in mentoring some of them. Just contact me for details.
These days there are tons of new projects and users/contributors have little patience to look into its documentation and decide whether the project is worth their time.
A well written, clear README can make a big difference for a project popularity.
The idea is to use the number of stars as a proxy for README quality and train a classifier than can predict whether a README will be well-received or not. (The number of stars is a weak signal for just the README, but it might be enough.)
With the classifier in hand, it might be possible to generate edits recommendations, similar to recommendations on how to write a good bug report.