Spotify + Machine Learning. A game changer.
The issue of taste.
How many songs exist today? Though there’s no consensus, the order of magnitude is estimated to be in the hundreds of millions. Added to this stock are the thousands of songs released each year. Due to this sheer volume of music, listeners are challenged to discover music they like. Additionally, some listeners don’t know exactly why they like a particular song and may even prefer a broad range of genres. This is what makes stumbling upon a song or getting a recommendation from a friend so exciting. Song discovery has historically been aided by subjective sources such as DJs. In the 2000s, music streaming platforms such as Pandora relied on manual curation or tagging to drive their song recommendations.1 Though better than discovering songs by pure luck, discovery aided by manual curation and tagging is ultimately tough to scale and can’t provide truly individualized recommendations.
Spotify… A game changer in music streaming industry.
Launched in 2008, Spotify is the world’s largest music streaming service with 159 million monthly active users across 61 countries.2 At the time of the company’s initial public offering (IPO) in April 2018, Spotify generated €4 billion in revenue and was growing 45% annually. Music streaming services have experienced outsized growth compared to the music industry overall (see Figure 1). Accompanying this rapid growth is intensifying competition as Pandora, Apple Music, Tidal, SoundCloud, Amazon, and Google all fight to attract new subscribers.
How Spotify winning the market drastically.
Core to Spotify’s strategy for winning in this crowded market is its ability to provide personalized recommendations and help users discover new music, which is enabled by its investments in machine learning. In its IPO prospectus, the company highlighted this strategy stating that it will, “continue to invest in our artificial intelligence and machine learning capabilities to deepen the personalized experience that we offer to all of our Users” and that “this personalized experience is a key competitive advantage.” Given Spotify’s deep pool of data (200 petabytes compared to Netflix’s 60 petabytes)2, the company is well-poised to create competitive advantage and provide users with a continually improving service.
Spotify’s strategy for recommendation.
Spotify’s strategy has consistently focused on machine learning. First, its machine-generated, personalized playlists such as Discover Weekly and Release Radar account for 31% of all listening on the platform compared to less than 20% two years ago. The company employs three types of machine learning to enhance its recommendation engine: collaborative filtering, natural language processing (NLP), and raw audio models1. Through collaborative filtering, Spotify provides recommendations to users based on the preferences of users with similar tastes. With NLP, the company scours articles, blogs, and song metadata to generate “tags” associated with each song and compares those tags with those of other songs. The company also analyzes which artists or songs are frequently mentioned along with the song in question to refine the pool of song recommendations. Through raw audio processing, Spotify is able to identify commonalities between songs through their musical elements (e.g. tempo, time signature, key). While collaborative filtering and NLP allow Spotify to point users to popular songs they may enjoy, raw audio processing allows the company to make predictive suggestions for songs with very little user awareness.
Spotify uses three forms of recommendation models :-
- Collaborative Filtering : Collaborative Filtering is a popular technique used by recommender systems to make automated predictions about the preferences of users, based on the preference of other similar users.
- On Spotify, the collaborative filtering algorithm compares multiple user-created playlists that have the songs that users have listened to. The algorithm then combs those playlists to look at other songs that appear in the playlists and recommends those songs.
- Natural Language Processing(NLP) :- NLP is the ability of an algorithm to understand speech and text in real-time. Spotify’s NLP constantly trawls the web to find articles, blog posts, or any other text about music, to come up with a profile for each song.
- Convolutional Neural Network :- Convolutional Neural Networks are used to hone the recommendation system and to increase accuracy because less-popular songs might be neglected by the other models. The CNN model ensures that obscure and new songs are considered.
- The CNN model is most popularly used for facial recognition, and Spotify has configured the same model for audio files. Each song is converted into a raw audio file as a waveform. These waveforms are processed by the CNN and is assigned key parameters such as beats per minute, loudness, major/minor key and so on. Spotify then tries to match similar songs that have the same parameters as the songs their listeners like listening to.
- With these key machine learning models, Spotify is able to tailor a unique playlist of music that surprises its listeners every week with songs they would have never found otherwise.
Second, Spotify has bolstered its strategy through several acquisitions. In 2014, Spotify acquired Echo Nest at a $100 million valuation3 strengthening its music recommendation capabilities. In March 2017, Spotify purchased Sonalytic which develops audio feature detection technology.4 In May 2017, Spotify acquired Niland, a startup which provides more accurate music search and recommendation.5
Finally, Spotify is exploring the use of machine learning to help artists compose songs. To do this, Spotify hired François Pachet in the summer of 2017 to be the Director of the company’s Creator Technology Research Lab. Though Pachet views machine learning as a complement to the artists’ creative process, one could envision a future where Spotify uses its machine learning capabilities to compose its own original content based on its users’ preferences.
Though it seems like Spotify is the market leader today in part due to its use of machine learning, how else can the company ensure that no other company sprouts up with a better algorithm to make personalized recommendations? How will Spotify, given its market clout, shape artists’ process of new music creation?