Spotify is one of the world's largest legal streaming music service platforms. In the process of product development, most of the users' favorites are attributed to its music recommendation system, which allows users to listen to their favorite songs every time. So how does Spotify know about users? Today I will use the super product strategy methodology to analyze for everyone.
For Spotify users, it has long been a habit to receive Spotify’s Discover Weekly every week, which includes 30 different styles of songs that users have never heard before. The magic is that every song can make users feel satisfaction.
"Spotify is my favorite music application, especially Discover Weekly. I think it knows me so well and knows my music taste better than anyone. The music recommended every week makes me very satisfied. I don't even feel it. As long as it exists, I feel that it is everywhere." This is what a loyal Spotify user said.
According to data, most Spotify users are addicted to Discover Weekly. It has a large number of fans. Even Spotify has a feature that is too popular and rethinks the business model of the company. The purpose of investing a lot of resources is to make weekly recommendations more Precise.
Since Spotify launched Discover Weekly in 2015, the number of users using Spotify has increased day by day. So how does Spotify accurately recommend the 30 music to each user?
Before that, we can compare Spotify with other music companies.
When it comes to the function of recommending music, everyone is no stranger to it. As early as 2000, Songza started to recommend music, but at that time it was recommended to users through manual screening of songs, and Songza would also invite people in the industry to make songs. single. However, it is unavoidable that those in the industry tend to prefer their own music taste when choosing a playlist, not based on the user's music taste.
With the passage of time, the former has made wrong demonstrations, while the latter has been improved. As an early company in the field of music recommendation, Pandora, it describes each song according to user keywords and puts these songs on the corresponding Tags, and then filter songs through codes, so that similar music forms a set of playlists to recommend to users.
At this moment, a music company The Echo Next from MIT Media Lab was born.
Its appearance has subverted the entire music recommendation field and will take a big step towards meeting the individual needs of users. Use algorithms to analyze the text and melody of songs, identify music, personalize recommendations, create playlists, and analyze music. Filter the songs that users may like.
In summary, we will find that the recommendation functions of these music software are becoming more and more complete. So how does Spotify build its own music recommendation system and stand out from the crowd?
In fact, Spotify does not only use a music recommendation method, it combines the recommendation model of other music software to create its own unique recommendation system.
The three super product strategy methodology behind Discover Weekly:
1) Analyze users and tap user needs
2) Process music source files and audio channels
3) Natural language processing, analysis of text
Analyzing user behavior and finding inspiration for user needs depends on Spotify’s competitor Netflix, which is the first company to use collaborative filtering technology to build a music recommendation system. They use a scoring system to understand users so that companies can recommend to users and their preferences. Similar content.
Since the company's success, almost all applications have adopted this technology.
Unlike Netflix, Spotify does not adopt a scoring system. Instead, it analyzes users with implicit feedback. For example, after the user listens to a song, Spotify will prompt whether it needs to be collected or the user is Have you browsed the homepage of the singer after listening to a song?
Judge the user’s preferences through these analyses. Assuming that two different users, A and C, three of the five songs they like are the same, then they are likely to be users with similar tastes. Therefore, they will have There is a high chance that you like songs that the other party likes but haven't heard.
Therefore, recommend songs that A likes but songs that C has not heard to C, and recommend songs that C likes and songs that A has not heard to A.
Through the above analysis, we will find that Spotify’s super product strategy only satisfies the most basic user needs and satisfies the user’s curiosity. At the same time, it finds the similarities between music and recommends users similar to their historical records. Music, let them hear the music they want.