In this liveProject, you’ll design a movie recommendation system step by step, from initial development all the way to a production-ready system. You’ll begin with preparing the data, then you’ll analyze the data, and finally, you’ll preprocess and export it. Along the way, you’ll gain a firm understanding of the data and your users, which is key for developing a system that makes appropriate recommendations. While this liveProject doesn’t contain any machine learning, going through these steps will prepare you for feature engineering and hyperparameter tuning later on.

Once you’ve handled the data, the real magic can begin! In this liveProject, you’ll implement a basic recommendation system using the TensorFlow Recommenders framework—designed specifically for this purpose. First, you’ll calculate four baselines that your future models will have to beat. Next, you’ll learn the basics of developing a model using TensorFlow Recommenders, then design a simple, two-feature Two Towers model. Lastly, you’ll enhance this simple model by adding all the features you created in the previous liveProject while maintaining model performance that beats your four established baselines.

Behind the scenes on websites like Amazon, Netflix, and Spotify, models make predictions on thousands of items every day. Then, based on what they’ve learned, they choose only the best recommendations to display for every individual user. In the real world, performing thousands of predictions one by one, as in a notebook-only model, would be highly inefficient. In this liveProject, you’ll reconfigure the models you implemented in the previous project to accept a list of items for each user and then evaluate all items at once—choosing the best recommendations much more quickly and efficiently.

Real-world recommendation systems work with millions of users and billions of items. To handle this massive scale, recommendation systems tend to be divided into two types of models (retrieval and ranking), running one after the other, narrowing down the set of items each time. In this liveProject, you’ll implement a retrieval model using TensorFlow Recommenders, combine it with a ranking model, then create a fully functional pipeline going through both models, using a Feature Store. Finally, you’ll explore the scenario where you can’t (or choose not to) run both retrieval and ranking models online in real-time, leveraging the Feature Store once more to use the retrieval model offline.

Since recommender systems train and learn over the data they recommended themselves, they will never train over, learn, or recommend items that they didn’t already recommend for some reason, such as insufficient ranking to be seen by the user. It’s important to break this Feedback Loop in order to ensure that suitable recommendations aren’t missed. But you must strike a balance between deviating (just enough) from the system’s predictions through exploration and not defeating the system’s purpose altogether. In this liveProject, you’ll learn three methods of exploration that help you provide better recommendations to your users, as well as the costs and benefits of each.