asder99 wrote:Oh, I'm sorry. Your other idea seem unapplicable too, because I constantly add new album to my collection... and I hate to spent too much time to classify it properly, so I just auto-tag them and everything it's done in 2 clicks.

Last.fm just analyze the singer and the songs you're listening to, and suggest you other similar artists and songs... check it out, it's not bad!

You can do it in not many more clicks by doing your autotagging and then just adding in a macro genre. What you have at the moment, based on your description of what you're doing, is an uncontrolled mess.
I did a bit more reading on Last.FM. I can't see that it is "analysing" songs at all. It's basing recommendations on song tags, social information and on your own behaviour.
Looks to be a bit like Amazon, which recommends stuff I might like, but it doesn't somehow analyse book text and authors to do so. It's looking at pre-existing classifications, my buying behaviour, and the buying behaviour of others (plus any manual amendments I make to the recommendations list).
According to
http://www.last.fm/about:Scrobbling helps us tell you what songs you play most often, which songs you like the most, how much you’ve played an artist over a certain amount of time, which of your friends have similar tastes… all kinds of stuff. By focusing on the music you already play we can help you discover more music.
and
Scrobbles mean we can deliver personalised recommendations for every single Last.fm listener, every single day. We compare what you play to the scrobbles of millions of listeners around the world, meaning your recommendations are the result of more than 43 billion scrobbles and counting.