Why are retrieval's outputs not used in ranking task?

[jack-pan-ai]

Hello, there is seemingly no use of retrieval's output when doing the ranking task. So, I'm wondering in your code, if I would like to use retrieval output when doing the ranking task, what should l do? Thanks a lot !!

[keivanipchihagh]

Here are the steps on how to use the output of the Retrieval for Ranking:

1. First give a user_id to the Retrieval model:

user_id = '41'
afinity_scores, movie_ids = scann_layer(
    tf.constant([user_id])
)

Which would output something like:

>> movie_ids[0, :5]
[b'196' b'969' b'211' b'215' b'174']

2. Then, pass the models_ids to the Ranking model:

for movie_id in movie_ids[0, :5]:
    _ = ranking_model({
        "user_id": np.array(user_id),
        "movie_id": np.array([movie_id])
    })
    print(_[0][0].numpy())

Which would output something like:

3.4644787,
3.3514086,
...

If you make this into a pipeline, you will then pass in a user_id and get movie_ids with their corresponding ranking scores.