Discoveries with Recommendation Systems

Have you ever watched a Netflix movie that becomes your favorite, based on Netflix’s recommendation system? Or discovering your favorite song from the Spotify recommendation list?

Sherlock has always been my favorite TV show. Then, I started the Blacklist TV series upon Netflix’s recommendation and was fascinated.

My Spotify playlist is too complex with many different songs from different genres. What is important to me in at song, is the harmony of the words and music, and it’s making me feel like I am in a parallel universe.

After Spotify recommended the duet of Mylene Farmer with Gary Jules, I discovered one of Mylene Farmer’s song named ‘Innamoramento’. I love this song.

What if all these suggestions didn’t exist? How many people in the world couldn’t discover their favorite song or watch their favorite movie ?

How do companies like Netflix, Amazon, Spotify use these systems and understand what we might like?

The answer to this question is actually just three words: Hybrid Recommendation Systems

Hybrid Recommender Systems use Collaborative Filtering Systems and Content-Based Filtering Systems together.

Collaborative Filtering Systems: Divided into two points of view. If your point of view is from the user’s point of view, it is considered user-based, and if it is from the product point of view, it is considered item-based.

User-based: Analyzes users’ behavior and provides recommendations based on the actions of other similar users. It works by creating a user-item matrix for items by users.

Item-based: Based on products purchased together or rating movies-songs etc. rated together high by users.

Content-Based Filtering Systems: Suggests similar products to you based on the product you consume, the movie you watch, or the song you listen to.

For example, if you’re a Marvel fan and you’re watching a movie from the Marvel universe and you’re seeing Marvel movie recommendations, it’s a Content-Based recommendation.

Let’s say the movies you watched, liked, and gave high ratings were “Interstellar”, “Joker” and “Back to the Future”. The system finds users who watch the same content as you and analyzes those who give high ratings to the same movies as you. The most similar users are filtered out and make suggestions to you based on the content of similar users. This is a User-Based Collaborative recommendation.

On the other hand, let’s say that Interstellar and Joker are highly rated by other users. If you’ve watched and enjoyed Joker but haven’t watched Interstellar yet, you’ll likely see Interstellar on your Netflix “You May Also Like” list. This is the Item-Based Collaboration recommendation.

Also, did you know that Netflix uses visual predictions in its recommendation system to impress their members? Yes, this is so crazy. Let’s say Netflix recommend Rush Hour to their member who likes crime and action movies. They can add the most action-packed scene from the Rush Hour movie as cover art to increase the possibility that the member will click.

Or are you a fan of The Matrix? When The Matrix 4 Resurrection will be released on Netflix, compare the cover photos on Netflix with your friends who have different tastes than you 🙂

Let’s do an exercise on what we talked about with you. Prepare a Recommendation System for user_id = 108170 with this Kaggle dataset (https://www.kaggle.com/hiraahmed/movielense20m).

You can find all the codes in my Github repo.

https://github.com/HalenurBulgu/HybridRecommendationSystem

Now is the time to listen to that new song you discovered on Spotify after reading this article 🙂

Stay tuned to see you on the next topic!