You have constructed several recommendation models, from content-based filters to matrix factorization. A natural and necessary next question is: how good are they? A model's effectiveness is not absolute; its performance depends on the specific goals of the application. To make informed decisions and improve your systems, you need a formal method for measuring and comparing their output.

This chapter introduces the techniques for that assessment. We will start by distinguishing between offline evaluation, which uses historical data, and online evaluation methods like A/B testing. The focus will be on the practical application of offline metrics, which allow you to iterate and test models before deployment.

You will learn to implement and interpret several industry-standard metrics for different evaluation tasks:

• Prediction Accuracy: For scenarios where predicting a user's exact rating is the goal, we will cover metrics like Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE).
• Ranking Quality: For most applications, the quality and order of the top recommended items are what matter most. We will explore a set of ranking metrics, including Precision and Recall at $k$, Mean Average Precision (MAP), and Normalized Discounted Cumulative Gain (NDCG).

By the end of this chapter, you will have a practical framework for quantifying the performance of your recommendation models, allowing you to compare different algorithms and tune their parameters effectively.