How to normalize small predictor values?

Question

data mining

I am implementing a recommender system with binary scores - 1 and 0. The score prediction formula, in accordance with the SVD algorithm:
r = mu + b_u + b_i + <factor product> ,
where b_u and b_i are the base predictors for user and item, respectively. The problem is that the base predictors turn out to be very small, with a few zeros after the decimal point, and do not affect the prediction of the score. As a result, the ratings for the user are the same. How to overcome this predicament?

Answer 1

How do you calculate the base predictors b_u and b_i?

Answer 2

Options -
1. Normalization - x=(x-mean(x))/std(x)
2. Principal Component Rotation (PCA).
3. Reversal to non-linear principal components - KPCA.
It will be like a data preprocessing step.
Normalization will make the influence of the variable independent of the quantitative measurement.