Predicting New Users with Missing Values

For content-based filtering, how can we score new users with limited features (e.g. null values or missing values)? Would we have to impute those features in order to get a prediction?


Hello Dan,

Gradient descent can’t handle null values, so you need to impute them.

There are at least two levels of considerations: first, will you build your model only on content that’s provided for sure; second, you may google about missing data techniques and apply any one of them, but the bottom line is that it has to perform satisfactorily on evaluation.