search

tensorflow / recommenders

TensorFlow Recommenders is a library for building recommender system models using TensorFlow.
Apache License 2.0
1.78k stars 267 forks source link

Any example for sequential recommendations? #393

Open jiwidi opened 2 years ago

jiwidi commented 2 years ago

Hi! Ive been reading the docs and I couldnt find any example for sequential recommendations, commonly used for session based recommenders.

Is that use case covered by the functionalities in this repo? Any example about it?

Thanks!

ydennisy commented 2 years ago

Hi @jiwidi I am looking for the same - best I found was this issue: https://github.com/tensorflow/recommenders/issues/119

jiwidi commented 2 years ago

@ydennisy Ill check it out thanks!

ydennisy commented 2 years ago

@maciejkula I would also like to suggest that I feel this could be quite a good tutorial to add if possible :)

kylemcmearty commented 2 years ago

Published 21 hours ago: https://github.com/tensorflow/recommenders/blob/main/docs/examples/sequential_retrieval.ipynb

Pretty version: https://www.tensorflow.org/recommenders/examples/sequential_retrieval

windmaple commented 2 years ago

Yep, we created that tutorial after seeing your feature request here

almirb commented 2 years ago

I read the TFRS paper.

How can we share the same movie embeddings for movie past watches and movie label? This approach is mentioned on a paper image: https://i.imgur.com/LABo21v.png

Using the same vocab for building the embeddings is enough? Thanks!

almirb commented 2 years ago

I read the TFRS paper.

How can we share the same movie embeddings for movie past watches and movie label? This approach is mentioned on a paper image: https://i.imgur.com/LABo21v.png

Using the same vocab for building the embeddings is enough? Thanks!

Maybe @maciejkula or @windmaple could shed some light on this.. :)

maciejkula commented 2 years ago

You will need to share the embedding table you use for past movie watches and movie labels in order to share the embeddings.

almirb commented 2 years ago

You will need to share the embedding table you use for past movie watches and movie labels in order to share the embeddings.

Can I share the same embedding table even if I need to use a "string splitter" as the first layer, and adding a GlobalAveragePooling1D as the last one?

Past watches needs a splitter... movie labels don't...

maciejkula commented 2 years ago

As long as you pass the ids into the same Python embedding layer object, you will be sharing embeddings, regardless of the processing.