What would be the shape of the movies data in the retrieval step when large number of movies

So lets say for 10m is 10 million records for movies and n_u is length of user features, J is total number of users in the platform at time T and n_m is the length of movie features and I is total number of meaning, which in this case is 10m.

In the following video at 2:53 secs

Do we retrieve the movie feature for each user or all users at once. Also what would be the shape of the ndarray that will be passed to items vector neural network in ranking step (the second model we train).

Let’s begin with a question to you. From the slide that you have shared, roughly how many movies do you think we can retrieve for a user?

From the screenshot I shared I interpret we take 10 movies for each user, then rank it using NN and show on the dashboard of platform. Also We can do take some similar user, sort the user by distance range and take that batch of similar users

Why go so far? From the slide, there are 3 steps, in the first step you can get 100 movies, right? What about the second and the third steps?

Umm… still not clear

Don’t worry. Let’s go through it together.


Do you see how it comes to 10 x 10 = 100?


What I am going to do is to walk through some lecture materials with you so that finally you can answer the questions yourself. It is not like one jump to the answer, instead, it is going to be more gentle and step-by-step, because in the way I want to find out and clear any gap between your understanding and the lectures. If you think going through it slower may help, then I will adjust. Any moment you think you have come up with an answer that you can explain and that I can understand, we can stop. I think you and I are working together on this, and I hope you think the same too. :smile:

If you are fine with the 10 x 10 =100, then we can move on.


Yes it would be 10x10 matrix