Member-only story
What Is Singular Value Decomposition? How It Helped Netflix Recommend Movies
Let’s understand SVD through movie ratings, linear algebra, and Python.
A movie recommendation is a prediction. Netflix has information about the movies we watch and rate, but it does not know in advance whether we will enjoy every movie in its catalogue. It has to use the available information to make a reasonable guess.
Suppose a viewer gives high ratings to romantic comedies and romantic dramas but low ratings to horror movies.
Another viewer has a similar pattern.
If the second viewer enjoys a particular romantic film, that may tell us something useful about the first viewer’s preferences.
This is the basic intuition behind collaborative filtering, a family of recommendation methods that uses patterns in people’s behaviour to suggest items they might like.
But how do we find these patterns when we have millions of viewers and thousands of movies?
One mathematical technique that helps us understand this problem is Singular Value Decomposition (SVD). It allows us to break a matrix into three matrices that reveal different aspects of its structure.
