User-Item Matrix
The core data structure of a recommender: users as rows, items as columns and ratings as values, mostly empty.
When to use
Known ratings are data and the empty cells are the predictions to make, so every recommender is a method for filling the blanks. Built in pandas with pivot_table(index="user_id", columns="item_id", values="rating"), which converts long format to matrix format. Its emptiness is Sparsity.
See DS Lec 6.