[Question] How best to re-use large, repeated text features.

Hi,

From reading the tutorials, the way data is passed into the model would mean lots of repetition, for example in the movie example, you would keep passing the same movie many times for different users. If we have a movie summary, this would mean having to store that summary many times over - which can get pretty big.

user | movie | movie summary

My question is what is the best way to create a "FeatureLookup" of sorts where for the same movie name / id we could grab the summary which would be stored once, and pass this in during training / inference?

tf.keras.Sequential([
  tf.keras.layers.experimental.preprocessing.StringLookup(
    vocabulary=unique_movies, mask_token=None
  ),
  tf.keras.layers.FeatureLookup(), # Fake layer, here as example, returns movie summary.
])

Thanks in advance!