For a recommender system, is there a real data matrix that is about 500 by 500 that is complete and has no missing entries?

Question

I was trying to get together a fully observed matrix for collaborative filtering. For example, some data matrix that was (more or less) at least 500 by 500, of say for example, 500 users that all rate the same 500 movies.

I tried getting something like this from the netflix data set and only got around 256 by 150, which seemed too small.

Is there a data set that is already processed like this that I can use? It doesn't necessarily need to be movies, but it would be best if it was a recommender system data set.

Or maybe a suggested data sat that has a high chance of getting a sufficiently large data matrix. The larger the better obviously.

Answer 1

Here is a listing of Recommendation and Ratings Public Data Sets For Machine Learning.

500x500 may be ambitious, so instead of movies you may have to use music (shorter content means more reviews)

In particular, take a look at (copy/pasted from site):

• Last.fm - Music Recommendation Data Sets

• Yahoo! - Movie, Music, and Images Ratings Data Sets

• Audioscrobbler - Music Recommendation Data Sets

• Amazon - Audio CD recommendations

Note that Last.fm also has an API