[heyplanet]

Of all the music recommendation systems out there, this one is my favorite:

http://www.gnoosic.com

It has an uber simple interface (you don't even need to log in) but no other system gets me as well as this one.

What surprises me in the given article is the graph on the last page. Why was the treatment group still listening more weeks after they already were getting the old recommendations again?

I wish the graph would reach out further into the past then just one week before the test. So one could get a feel for how much of the variance in the graph is random.

[laxatives]

Would be neat if it had a slider for exploration. All my answers were pretty dominated by really obscure and out music. Sometimes I really want that, some times I want something a lot accessible I can just hit play on without my full attention.

[ganzuul]

Their 6th reference mentions Primacy and Novelty as an explanation for the lasting effect, but it is not clear to me that the reference exaple and the current research are directly comparable.

Seems that the 95% confidence interval represents the variance of previous data.

[ep103]

This is literally the only recommendation engine I've ever had any success with. IIRC it was written by someone on HN

[jacobobryant]

I'm working on a music recommender right now, I'd love your input: https://news.ycombinator.com/item?id=20584508 Mainly my work has focused on re-recommending songs you already know, e.g. in an internet radio kind of setting. I'm thinking of incorporating something like gnoosic for recommending new artists.