Can someone help me better understand the need for the additional component of Bias with an embedding vector? I understand the intuition of assigning a Bias to a user or movie, but couldn’t that just be another dimension added to the Embedding Vector?
My guess is it has something to do with it being added after the fact, but I don’t understand intuitively how summing post-dot product is all that different from just being included in the dot product.
Thanks all 