9.2 A note on sparsity of rating matrix

The user-item rating matrix is a sparse matrix because most people do not provide ratings. Think about how many times you provided any ratings on Amazon or Netflix. Therefore, most cells of a rating matrix have missing values.

A way to improve upon this situation is to use 1-0 rating scheme whereby if a user engages with an item (buys a book, watches a movie, listens to a song), we fill the cell with 1 and otherwise it is filled with 0. This still does not solve the problem because a 0 has more than one meaning. You may not watch a movie because you don’t like something about the movie (genre, actors) or you are simply not aware that the movie exists.

We will side-step this concern for the current exercise. In reality, you will have to find ways to address this because your recommender efficacy depends on it. Ideally you would like to recommend an item that the user is likely to appreciate and engage with.