No, but I've definitely thought of using sklearn's Naive Baye's implementation to give me paper highlights (a la Mendeley, minus Elsevier). You can train on paper abstracts you've previously found interesting/uninteresting.

Luckily arXiv RSS feeds are by category, so I scan the titles of all e.g. qbio papers for a day and add the interesting ones to my reading list.

ArXiv feeds are organized by category (more like tags in that papers could be cross-listed, but need a primary category). Daily traffic in different categories is very different. Eg, at some point I used to subscribe to hep-ph/hep-th/gr-qc and they combined to something between 50-100 papers a day. I’d mostly just scan the titles/abstracts, to pull out particularly interesting papers.