I have my RSS reader YOShInOn which subscribes to about 110 feeds. Over the course of 2 days it could gather about 7000 articles and choose 300 to show to me, it works as a content-based recommender (based only on my judgements) although it builds on a pre-trained BERT model. It does clustering over the latent space so I see a variety of articles, not just arXiv papers. Anything you see here
https://news.ycombinator.com/submitted?id=PaulHoule
was chosen by me out of the feed it shows me.
I have another project to build something for filtering streams of images that is sharing a lot of code w/ YOShInOn and may eventually get merged. One of the first things to merge is a bookmark manager which would treat bookmark items like things that come in out of the feed.
Hypothetically I'd like to have a really good search system but now I am (1) using a simple system that seems "good enough" and (2) not particularly looking for history in it.
Another thing immediately in front of me is I get a lot of articles like
https://phys.org/news/2024-02-kink-proton-spectrum-knowledge...
and for an article like that I might look up the paper referenced in it and make a decision that I want to submit that article or the original paper based on how obnoxious the ads are, is the paper open access, etc. There are other ways to do it but the master plan seems to be to scan the ScienceX article, ingest the papers, and construct a graph structure about how they are related.
