Using contextual features of the current user's page.

I am a little confused by the two towers architecture on what features should I use in each model (query & candidate).

I'm trying to build a recommender for consumer products.

I want the model to recommend products for new visitors between the training operations and consider the existing visitors' behaviour between these training operations. Thus I replaced the user_id with a sequence of the last X product visits per #119 .

I have at my disposal a lot of contextual information which I want to use. If I have got it right from Taking advantage of context features, I should use them only in the candidate model since they are product's and not visitor's attributes, right?

Based on the above, my current model is like:

class UserModel(tfrs.Model, ABC):
    def __init__(
        self, lookups: preprocessing.Lookups, hparams: preprocessing.RunParams
    ):
        self.past_visits_embedding = tf.keras.Sequential(...)

    def call(self, inputs: Dict[Text, tf.Tensor], training=None, mask=None):
        return self.past_visits_embedding(inputs["past_visits"])


class ProductModel(tfrs.Model, ABC):
    def __init__(
        self, lookups: preprocessing.Lookups, hparams: preprocessing.RunParams
    ):
        self.url_id_embedding = tf.keras.Sequential(...)
        self.category_embedding = tf.keras.Sequential(...)
        self.title_embedding = tf.keras.Sequential(...)
        self.price_embedding = tf.keras.Sequential(...)

    def call(self, inputs: Dict[Text, tf.Tensor], training=None, mask=None):
        return tf.concat(...)

Where I'm really struggling is, can I use contextual features of the visitor's current page to infer recommendations for that page?
In other words, if a user is on product page A, can I use the title, category, price etc., of product A as embeddings in the user model to infer the next product?
e.g

class UserModel(tfrs.Model, ABC):
    def __init__(
        self, lookups: preprocessing.Lookups, hparams: preprocessing.RunParams
    ):
        self.past_visits_embedding = tf.keras.Sequential(...)
        self.current_url_id_embedding = tf.keras.Sequential(...)
        self.current_category_embedding = tf.keras.Sequential(...)
        self.current_title_embedding = tf.keras.Sequential(...)
        self.current_price_embedding = tf.keras.Sequential(...)

    def call(self, inputs: Dict[Text, tf.Tensor], training=None, mask=None):
        return tf.concat(...)

I really appreciate any help you can provide.