waffle

Amazon became the world’s biggest Internet retailer in part by making it so fucking easy to follow the breadcrumbs from everyone else. (Other buyers also bought; Visitors to this page who actually bought something ended up buying; Your weird neighbor, the one with the twitch and the “precious” bloodied butter knife, bought… You know the drill.) Netflix sees the value in them too and has initiated a prize for improving on their algorithms in terms of accuracy and speed. So one wonders why, three years following its introduction, the “Just for You” feature in the iTunes Store is so crappy.

Just for You has some potential right out of the gates. It ambushes you on the Store’s front page and recommends music based on the music you have already bought. And supposedly, it can also be “trained”. Just for You doles out a bunch of pages of recommendations; on each page is an album’s worth of assorted songs in the track list along with six albums. You see what you bought that makes iTunes recommend this album to you, you can train the recommendation beyond buying by saying “I already own this” or “I don’t like this”, which admittedly conserves precious cash.

This is a good start, but it leaves me wanting:

• The albums are the only things I can trace and train. I don’t have a clue why I’d like the songs.
• At least for me, there only ever seems to be about three or four pages of recommendations. This leads me to believe that the list of related items is pre-generated and not dynamic. This may be fine for the “listeners also bought” feature on albums, but not so much here on a more personal feature.
• There’s a very clear limit to how smart the recommendation engine is. For artists where I own several albums, it keeps recommending albums I already own, or it recommends the EP singles of tracks that I already own, regardless of if I own the “B-side” track or not.
• There’s no big lever I can pull to initiate a recommendation mode: “I wish to primarily get suggestions of tracks by artists INSIDE my corpus or OUTSIDE my corpus”. Along with a “really, I don’t like anything this artist does” button and perhaps along with ranked recommendations, say that albums by five separate artists recommend this other album, this could help.
• As long as there’s plenty of information on exactly how and as long as it is opt-in, I wouldn’t mind allowing iTunes to upload metadata from my library to swiftly and easily train it.

Just for You is a good idea. It’s a hub for recommendations, such as that you won’t have to hunt down your favorite albums to check their recommendations for new entries every now and then. This could be — and ought to be — a place where you’d be stuck for hours pulling the lever, seeing what the celestial all-knowing jukebox thinks you’d like. But today, its generated finite output, crappy recommendations and need for ages of potty training with all the data it needs available in the same damn program stops this from happening, and that sucks for everyone involved.