search

tensorflow / recommenders

TensorFlow Recommenders is a library for building recommender system models using TensorFlow.
Apache License 2.0
1.84k stars 276 forks source link

Has anyone implemented retrieval model in ways other than 2 tower model? #436

Open amit-timalsina opened 2 years ago

amit-timalsina commented 2 years ago

I have seen most of the people are implementing just the 2 tower architecture for the retrieval model. I wanted to try other architecture. So, if you have implemented or have an idea about ways to do that, please help me.

Also, Like for this implementation, how can we use TensorFlow recommenders? image

patrickorlando commented 2 years ago

hey @amit-timalsina, the picture above describes the serving case for a two tower retrieval model.

amit-timalsina commented 2 years ago

hey @amit-timalsina, the picture above describes the serving case for a two-tower retrieval model.

How, can you please elaborate?

AzizIlyosov commented 2 years ago

@amit-timalsina I guess, that model is implemented in softmax model section of this colab

maciejkula commented 2 years ago

The softmax model you picture here is a degenerate case of a two-tower model. The softmax is exactly equivalent to having a candidate tower with a single embedding layer, with one embedding vector per candidate.