By Samantha Howard

Let me first begin by clarifying that this is not an article endorsing Spotify. This is an article about how your own personal algorithms are a special code or recipe that is created by you, but Spotify just knows how to record it. When the program starts to understand your music taste, it develops a formula that only aligns with you, and that’s pretty bloody magic.

There’s over 100 million people every week who can access their very own Discover Weekly, which is a special delivery to you personally, giving you a serving of music scientifically created by you. Except, you’ve never heard it before, and the playlist is unique, created by an algorithm.

Sophia Ciocca reports in her piece, ‘’How Does Spotify Know You So Well?’ on some of the technological advancements in music app history which usually generates ‘ideal playlists’ based on a method of human curation. Songza was the first musical powerhouse to deliver this technology, but a human will be bias, and can only assume the tunes you will like. The system doesn’t quite know you well enough.

It was then when the musical intelligence agency from the MIT (Massachusetts Institute of Technology) Media Lab called The Echo Nest created the technology to analyse the ‘audio and textual content of music’. This process involves music identification, recommendation and analysis. This musical intelligence agency helps Spotify curate music recommendations which are personalised and driven by algorithms.

Spotify has three models for this process. Firstly, there are Collaborative Filtering Models — last.fm was the original developers of this technology, which involved collaborative filtering to learn from its users. Think of it like sharing recommendations between people which a robot can then learn from.

The formula for this process is a bit hekkas but looks like this:

Source: ‘How Does Spotify Know You So Well’ (online). Available at https://fivethirtyeight.com/features/spotify-knows-me-better-than-i-know-myself/ 

The second is Natural Language Processing (NLP) model. This model picks up words — but music is music I hear you thinking … how do words give the program the ability to track your music? Spotify has the technology to send web crawlers which search for WHAT PEOPLE ARE SAYING ABOUT SONGS. It picks up adjectives and language used to reference certain songs or artists and the other artists and songs spoken in addition to it too.

The third model that Spotify uses is called Raw Audio Models. Rather than taking recommendations, Raw Audio Models actually analyse the data of songs that may not have been recorded as good tunes by Collaborative Filtering Models or NLP. They use an even more special technology called Convolutional Neural Networks which identify similarities in tempo, key, time signature and mode.

When using all three of these together, Spotify Discover Weekly or Song/Playlist Radio is basically delivering you new music babies that you will love and kept in the womb — although you technically might not have known you were pregnant yet — and you didn’t have a baby bump.

Which leads me into how you can make music love children with your friends. Start a monthly playlist. Call it whatever you want, but regularly listen to Discover Weekly, or old playlists and save your favourite tunes. At the end of your playlist, you will reach an automatic play of Playlist Radio. You can also try song radio for ultimate bangers. Keep saving those tunes you love.

Get your friends onto monthly playlists. In their presence, when they play an ultimate bloody banger, save that to your playlist. Gradually, you’ll pick up what’s your vibe, and what’s their’s — and without meaning to sound corny-as-hell, you will quickly discover what makes your music taste so special and what makes their’s too.

Spread the music love.

Image credit: https://unsplash.com/photos/PDX_a_82obo