When I find an article or blog post I like, some sites and most blogs allow me to subscribe to a “feed” (implemented with RSS or Atom), so that I can receive more articles and blog posts. The feed is better than an email subscription because I control the subscription. With email, the sender has a list of who subscribe; with feeds, I have a list of what I subscribe to. Leaving this up to the sender means a spammer can subscribe me without permission (and this happens to me all the time). It means someone can put in the wrong email address, and I end up receiving their email (this happened to me recently, and I started receiving email about my being in the 27th week of pregnancy). With feeds these things don't happen, because I control the list.

I think the bigger promise with feeds is the possibility to subscribe in new ways. Right now, I can subscribe to a site's feed. If I see an post I like, I can subscribe to receive other posts from the same site. This assumes that if I like one post from the site, I'll like other posts from that site. Sometimes this is true; sometimes it isn't, especially when the author mixes many topics in the same blog (as I do). If I've marked a set of posts (A) I like, there are many ways to guess what other posts (B) I'll like:

  • Posts textually similar, but not identical to, to the ones I marked: words(A) and words(B) overlap, but not completely
  • Posts that link to the same articles as the posts I marked: X ∈ (links(A) ∩ links(B)) but X is not commonly linked to
  • Posts that are linked to by the same pages as the posts I marked: for many pages X, A ∈ links(X) and B ∈ links(X)
  • Posts that share the same tags as the posts I marked: tags(A) and tags(B) overlap
  • Posts that are marked by the same people: my set A and your set A overlap, so tell me which posts didn't overlap

There's some remixing going on already with Feedburner, Google Reader, and others, but I want to see more automated creation of feeds based on sets of posts as the input, rather than feeds as the input. To build the kinds of things I'd like requires someone to analyze text, tags, and links for every post. It's a lot of work, which is probably why it hasn't been done.