I found an enlightening dissection of the Pitchfork music articles (via travelers diagram). The author has built a MySQL database of the 100,000 different words used in all 5,575 reviews. By calculating scores for words used in positive reviews he tries to find out which kind of music the critics like. He found very positive scores for words like "sad", "dark", "crescendo", "massive", "distorted", "lush" etc. In the end he even composed two songs taking into account his findings. The first one Kissing God is ok but the whole is definitely less than the sum of its parts. The song is not cohesive, the elements only form a conglomerate. It comes as a kind of relief that the recipe to satisfy the critics doesn't seem to work. I think people listening to a lot of different music want to be surprised, they want to discover new sounds and they crave for a personal, individual touch of the music. All this cannot be achieved using a program even if it is as well thought out as this one.