Building a movie recommendation system
To build a movie recommendation system, we need some kind of user-movie interaction dataset. Fortunately, we can use the MovieLens 100k dataset (https://grouplens.org/datasets/movielens/100k/) provided by Grouplens (https://grouplens.org/). This data contains 100,000 movie ratings given by 1,000 users on 1,700 movies.
The following screenshot shows the first few rows of the dataset:
Figure 11.7 – The first few rows of the MovieLens dataset
From the preceding screenshot, you can see that we have user_id, item_id (movies), and the rating the user gave to the item (movie). With just this interaction and the use of embeddings, we can efficiently model the behavior of users and, as such, learn what kind of movies they like.
To understand how we'll build and learn this interaction with embeddings and a neural network, please refer to the following architecture diagram:
Figure 11.8 –...
