[arek2]

76k micro-genres seems much. For my website http://5000best.com/movies/ I created 40 main genres using IMDb tags together with the 100 million ratings from the Netflix Prize data (I was 43rd in that competition).

Additionally, earlier I extracted and named 12 new genres (those ones on the right) from the Netflix ratings alone - I described the process here: http://arek-paterek.com/book/predict_sample.pdf

[KaoruAoiShiho]

What do you mean much? When working like this the more the better. There's an opportunity for an open source variant of this technology.