How to have user history in the query model

TensorFlow General Discussion
1

I have a two-tower model. Query model has info about the users such as user id and gender, and candidate model has item-related information item id, description etc.
How can I add user’s history (what item they viewed before) to the query model? How to integrate the item embeddings as history to the query model?

3

Hi @gulcek,

You can integrate the user’s history to query model. Here is the sample code to generate user history along item related information.

  • Get the embeddings from Candidate model by passing embeddings.
  • Aggregate the historical embeddings.
  • Concatenate historical embedding with user data.
    embeddings = candidate_model(item_ids)
    agg_hist = tf.reduce_mean(embeddings, axis =1)
    comb_data = tf.concat([user_features, agg_hist], axis = -1)
    Finally fed this Concatenated data to your query model.

Thank You

1 Like