Twitter’s biggest problem is that Twitter increases or decreases in value along with user’s willingness to curate their feed. Many will never do that. So Twitter tries to do the work for you, but guessing exactly who/what you’d be interested in without tons of info is virtually impossible. The closest they can get is the current logged-out homepage: “Here are a bunch of random categories. You like the NBA? Maybe cele…
That was my intuition as well. But it turns out that you can generate good recommendations without tons of info -- as long as you are selective about what signals you use.
We learned this firsthand when building https://recent.io/ for news recommendations -- the backend is able to generate relevant suggestions that appear in the app after users have read an average of 5 articles. That's based on real-world user testing.
If Twitter recognizes repeat visits by people who don't (yet) have accounts, it should be able to try a similar approach. On the other hand it took a while for us to get it right.