Collaborative Filtering: Why not Linear Regression

[Louis_Farenci]

I just got finished with the Collaborative Filtering Algorithm video. I am a little confused as to why linear regression cannot be used to learn parameters. What I mean is that if you are given w and b and ratings (y), isn’t optimizing for x just the same process with (basically) the same cost function to minimize?

For the movie-rating example, if you are given parameters w and b for one user, why is it that you cannot learn features for the movie with linear regression in the way I mentioned above? Is it merely because it will not generalize well to new users?

[TMosh]

Yes collaborative filtering can be viewed as linear regression.

The difference though is that the training data is in the output, and it’s a matrix of users and ratings. You’re trying to learn both the weights and features that give the best matching model.