Free Step-by-step Guide To Become A Data Scientist
Subscribe and get this detailed guide absolutely FREE
Ever asked your friends or friends of friends and their friends for music recommendations? Well, I have. Because, I like music. A lot. So, I’m always looking for new tones to hum or songs to sing about how I feel in any moment. Unfortunately, Spotify is “unavailable” in India. And, I am forced to rely one of the age-old recommendation systems – my social network. If only I could Spotify!
Discover Weekly Recommendation Systems
Every Monday, over 100 million Spotify users receive a carefully crafted custom playlist of 30 songs that most suit their musical tastes and preferences. How does Spotify make accurate predictions regarding user tastes and likes? By using a combination of these three recommendation systems.
Recommendation System 1: Collaborative Filtering
I’m sure you have heard of the saying, “birds of a feather flock together”. Well, they listen to the same music too! Collaborative filtering uses the underlying logic of the simple saying to predict Shyam likes Lotuses if he shares his liking for Daises, Roses and Lilies with Ram, who also happens to like Lotuses.
Collaborative filtering is one of the most common recommendation systems. LinkedIn uses it to recommend the “people you may know” to help you add them to your network. Again, the logic being that if Ram and Leela know Ganesh, Ramesh and Mahesh, then chances are Leela will know Kalpesh if Ram knows him.
Netflix uses the technique to recommend movies. They do it effectively by collecting ratings from their users and actively profiling their tastes to match them with the tastes of similar users.
How Spotify Uses Collaborative Filtering
Spotify recommends music by implicitly tracking user behavior. They record the number of times a user listens to a song, what songs they add to their playlists and what artists they demonstrate an interest in by visiting their pages.
Additionally, they also profile each of their songs by observing, who is listening to them and how much. They combine the knowledge gained from user profiles and song profiles to make highly accurate predictions.
Spotify’s ‘Discover Weekly’ playlists are extremely popular among users because they are tailor-made according to user tastes and interests. Spotify makes its users feel like they are known every Monday by speaking to them in musical tones and voices that they didn’t even know they wanted to hear.
Recommendation System 2: Natural Language Processing (NLP)
It’s a curious thing how languages can be so efficient in communicating even the most abstract of ideas and emotions. What we know is that all forms of communication involve two parties – the communicator and the listener or interpreter. When machines play the role of the listener, we refer to them as processors. And they’re quickly becoming proficient in the task of processing language with machine learning.
Spotify uses Natural Language Processing – one of the most in-demand machine learning skills – to learn, once again, their tastes and interests. Humans say a lot on digital platforms. Every comment, every post, every word used to describe songs and how much we like or dislike them is information that machines can now process and interpret for meaningful purposes like improving recommendations.
Recommendation System 3: Audio Processing
If the first two recommendation systems weren’t enough – and they weren’t as you will soon discover – Spotify seals the quality of their recommendations with this third system, which gets to the heart of the matter.
Songs are aural. So, the best way to compare one song with another is by paying attention to what they sound like. Machines can now process audios using convolutional neural networks to extract details like tone, beats, tempo, etc that are proving to be extremely useful in identifying similar sounding songs and making recommendations more effective. It’s the same technique used in facial recognition, but instead of applying it to pixels Spotify applies it to sounds.
This third system is critical because it helps Spotify recommend new songs from not-so-famous artists that the first two recommendation systems don’t discover. These are songs that need promoting. They escape the first two recommendation systems as they are less known (unlikely to be many users’ playlists) and therefore also less spoken or written of.
Spotify attracts artists to their platform by recommending their songs to the most suitable audience. Audio processing is beneficial for artists and users. It is also beneficial for Spotify to discover new or lesser known music.
The recommendation systems that Spotify use collect and process large volumes of data. With over 100 million users and 30 million songs on their database, it’s remarkable that they can maintain data pipelines that allow them to actively personalize recommendations for every user every week. Of course, advancements in big data technologies have helped and made this task easier. Spotify uses Hadoop clusters to distribute the load of managing such voluminous data and discover new songs weekly.