medium10 pts · first answer countsembeddingsfeature-engineering

Millions of product IDs as a feature

A recommendation model needs to use "product_id", which has 5 million distinct values. What is a practical way to represent it in a neural network?

A.One-hot encode it into 5 million columns stored denselyB.Learn a dense embedding vector per ID (optionally with hashing for rare or new IDs)C.Use the raw integer ID as a numeric featureD.Drop it, since IDs never carry useful signal

Discussion 0

Join free to post in the discussion.

Millions of product IDs as a feature

Discussion 0

Related questions