Hi James,
Manish Barnwal

Hey Manish,

If your item features are strictly metadata, then there is no restriction. If you add an ‘indicator variable’ to your item features, then the item would have to be in the training data so an embedding for that indicator variable feature can be learned.

You can read more about indicator variables’ impact on the model in section 2.3 of the paper: https://arxiv.org/pdf/1507.08439.pdf

Hope that helps!


One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.