I've worked on various music recommendation systems just for my own private consumption for 15 years. My current approach at least makes me really happy. The p

I've worked on various music recommendation systems just for my own private consumption for 15 years. My current approach at least makes me really happy. The problem is I'm likely radically different. The general problem is so are many people. But all in different, incompatible ways. For instance, my (current) approach is fundamentally label based. Things are organized by label and I "rate" releases as I listen to them. This leads to a generalized "label rating" which can be partially extended to (1) a rating of the artists who released on the label along with (2) albums and eps which may see a release under another label. Then there is a similarity index to other labels based on a percentage of artists who release on both labels and this leads to something going into a big "consider" pile which then gets treated randomly. It's not really a "recommendation" as much as it is a "might be related? i dunno" system. For people who want familiar songs they can sing along with, this is a terrible no-good system. For people who like to listen to the same song multiple times a day, this is equally useless. For those who want steadfast genre uniformity and things to sound more or less the same, this is also useless. This is designed to be exclusively long-tail and stay firmly within the confines way out at the end of the tail. But that's exactly where I want to be. In the pareto principle, everyone in the 80% looks more or less the same but everyone in the 20% is strange in a different kind of way. There is no 1 recommendation system that works for all customers. Users have to be segmented. It's utterly inescapable. That's why these systems try to just service the 80% and just not put any effort to the other 20%. The problem with this approach is the 80% is no longer 80%. More people are becoming snowflake 20%ers and this trend is likely to continue. The mainstream orientation engines of radio and cable television aren't influencing like they used to and the 'service the 80' approach just won't work any more. It's over. The single recommendation system approach is being tweaked to try to service this coalition that no longer exists. So what happens? Everything starts to suck. Google results, youtube recommendations, spotify, they all start to be unfocused, scattered messes that just discard all the signal, amplify the noise and return piles of nearly identical derivative garbage. Socially we've moved beyond the assumption that simply everyone in society will be in some conventional nuclear family with the wife vacuuming at home and the daughter playing with dolls but technologically we still seem to model humans using utterly conventional and outdated single purpose models that simply do not apply and are increasingly a poor fit for everyone. The new normal is that there isn't one and it's time for tech to play catch up.