Personalized feed

Several mechanisms contribute to the creation of a personalized feed on Spotify:

​

• Explicit and  Implicit feedback 

 

Explicit feedback is based on explicit input by users, usually content that the user has actually liked, rated or saved to his/her personal library while implicit feedback is based on inferences and predictions extracted from users behavior.

 

• Natural language processing (NLP): Spotify crawls the web constantly looking for blog posts and other written text about music to figure out what people are saying about specific artists and songs  which adjectives and what particular language is frequently used in reference to those artists and songs, and which other artists and songs are also being discussed alongside them. They subsequently compute the probability for an artist or song to be associated with a specific word.

 

• Audio models: They are based on the comparison of music audios - raw audio models don’t discriminate between new tracks and popular tracks

 

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

Audio features of Thriller by Michael Jackson

​

​

• Content-based filtering : Spotify Radio is an automated radio channel that is generated based on the comparison of features that describe tracks in a meaningful way (also known as content-based filtering) such as:

• Song

• Artist

• Album, etc...

 

• Collaborative filtering: This is based on the consumption of music by other people with similar music tastes. It uses a huge matrix that associates all users to all songs available on Spotify, as well as a map indicating the profile of different users of Spotify - also known as the “nearest neighbor” vector model (the distance representing the level of similarity among profiles). However, this mechanism is not fully effective because of the cold-start effect which is the impossibility to treat a specific product (here a track) due to a lack of sufficient data on it.

 

​

​

 

Generic feed

In addition to the personalized feed that is related to the music listened to by Spotify users, the app proposes a generic music feed that is not related to consumers’ personalized tastes and is seemingly the same for all consumers.

 

• Rankings: These are based on the most streamed songs in specific geographic areas

​

​

​

​

​

​

​

​

​

​

​

​

​
 

• New songs: A seemingly random selection of latest songs that are not related to users’ consumption but rather popularity.

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

• Genres & Ambiance: Based on a more or less conventional categorization of music:

  • in different music genres (jazz, RnB, Metal…) which is not always meaningful since there can be overlaps (some songs can be categorized in different music genres)

  • In situations (Diner, Romance, Fitness…)

 

​

​

​

​

​

​

​

​

​

​

​

​

​

​

 

Promotion of mainstream artists

In addition to a generic feed of songs that are not related to users’ taste and a personalized feed that is determined by several automatized mechanisms, there might be something more that is related to music promotion by record labels and would permit them to influence the supply of music on the platform to increase their streaming numbers and generate more profit.

Even though there has not been extensive research on the actual mechanisms that allow some artists to be “pushed” on the platform, regardless of users tastes, the saturation of the music offer by some artists is a rather clear indicator of a lack of neutrality of Spotify regarding the most mainstream artists.

 

• Drake’s Scorpion release controversy

 

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

One of the most eloquent examples of this phenomenon occured when Hip-Hop artist Drake released his latest album untitled Scorpion on June 29th 2018.

Premium Spotify users found themselves overwhelmed by the promotion of the album on the platform and the omnipresence of Drake’s music in playlists that were not related to their personal taste.

According to Billboard.com, “Spotify spent the weekend spreading the news with a first-of-its-kind promotional takeover that might have helped hype up fans but also turned off some subscribers enough to ask for a refund”.

 

This example clearly illustrates the ability of a supposedly neutral and personalized online platform to homogenize the offer of a service, even for a limited amount of time.

​

​