Mean noralization with binary matrix

I just watched a video about mean normalization for the implementation tip.
In the video, the user-item matrix is a 1-5 scale rating, and thus applying mean normalization by subtracting the mean of each row seems reasonable.

But what if that user-item matrix is binary (1, 0, and ?)??
Can we still apply mean normalization or Should we apply any normalization to each element in the matrix???

I think this thread is a duplicate.

I posted on the wrong topic last night. I have flagged it to be deleted.