Recommendation engines across digital entertainment services and social media broadly use two design approaches.
The first one is content-based recommendations, and the second one is called collaborative filtering.
Any movie/series we watch across any platform has several variables.
Let’s say you love the movie The Departed.
One variable is the category/genre — this one would fit into, say, Crime, Thriller, Action, or maybe Drama.
And then comes the star cast (Leonardo DiCaprio, Matt Damon, Jack Nicholson, Vera Farmiga, etc.) Martin Scorsese (the director).
Another variable could be the period when it was released – the mid-2000s.
One more could be the storyline — based on gangsters, for example.
Now, the content-based recommendation design would curate options on your home screen based on the variables we listed above.
The system starts with in-depth information about a title’s characteristics and then searches for other titles with similar qualities.
Now, the more you watch, the more input you provide to the content-based engine to sharpen its offering.
The second approach—collaborative filtering—is simpler, and we often see it as “people who liked this also liked…”.
The collaborative filters don’t have deep knowledge about the product, but they work by bucketing people into groups based on underlying tastes or personality.
But these are dynamic groups — you are moved from one group to the other based on your viewing pattern.
Again the more you watch, the more information the algorithm will collect and more precisely, it will place you in a group, and better its recommendations would get over time.
Netflix, like Spotify, combines both the recommendation engines to curate your recommendations.
And it is still easier to categorise movies — music is trickier.
Imagine an algorithm sifting through information such as “a dub production, a reggae feel, acoustic rhythm piano, use of a string ensemble and major key tonality.”
