.png)
As the AI and data industries develop, agencies actively utilize AI in their industries. In particular, platform agencies are developing in a new direction by utilizing AI. We will tell you how they are using AI in their business.
These days, playlists that curate songs that suit the situation are very popular! Spotify is a music streaming app and is famous for personalized music and artist recommendation services that are based on target users' preferences. There are tens of thousands of plies, including playlists that are appropriate for these situations, such as studying, picnicking, exercising, and driving, as well as playlists that make you feel like you're in a particular situation, such as an editorial shop playlist. Spotify introduced an innovative system in 2013 that recommends music based on AI, not a top chart. We'll tell you in detail how AI was introduced into the music recommendation system 😉
#1 Spotify: Big hit of the AI and Machine Learning
Spotify recommends 'Weekly Recommended Playlist' to users every Monday.
Spotify's users always said that this service is expensive, but they always recommend songs that users wanna listen to.
Spotify's music professionals can select songs and create a playlist, but it's practically impossible to choose a single song for all Spotify users, right? Spotify runs an algorithm to solve this problem.
They develop a music recommendation system for users and use machine learning in this process. When analyzing a user's online behavior, they use the Collaborative Filtering model, the Natural Language Processing (NLP) model, and the Audio Analysis model.
.png)
🤔 What is the collaborative filtering model?
💁 The collaborative filtering model also called CF and is the most commonly used model in the recommendation system. It also deals with user-item interaction data. Through this model, we can solve the question of "What kind of song would the person who liked this song like?".We can also expect that people with similar tastes have similar preferences for certain songs.
🤔 What is the natural language processing model?
💁 It is a model that analyzes people's online behavior (comments on news, blogs, etc., or SNS's likes, etc.).
🤔 What about the audio analysis model?
💁 It is a model that analyzes the genres, tunes, and keys of Spotify's music and classifies similar music.
Spotify uses these three technologies to create a giant matrix of two vectors (User Vector, Song Vector) that are composed of millions of data and to classify them. This is the secret to a personalized recommendation list!
Spotify's curation system integrates all of the information gathered through technology to create recommended playlists for individual users. We’ll tell you how the playlist is created.
First, they analyze the behavior data of listeners, such as selecting specific music or generating playlists. And then, they analyze web scraping and scoring to variable meaningful features and generate user vectors. Retained music generates variables representing individual characteristics such as genre, speed, and playback time, and matrices them. Finally, they select the main variables through factor analysis and generate song vectors.
*What is factor analysis? : It's a way to identify variables that have a substantial effect on the result value!
Next, they construct a matrix with millions of observations of the created user vector and the song vector as rows and columns, respectively. Then, a classification algorithm is applied to group the similarity vectors. Finally, you can configure a personalized recommendation playlist by adding songs that are missing from the group's common list to another user's recommendation list, or by analyzing new music properties and adding them to your personalized Weekly Playlist!
Spotify was able to create loyal customers thanks to these technologies. This is because it is a machine learning-based technology differentiated from existing music streaming companies that can recommend more perfect and personalized music to users. This immediately causes a lock-in effect. Even if I quit Spotify and joined another music app, I don't have any behavioral data that I've accumulated so far, so I like it! It's hard to get recommendations.
Collaborative filtering using behavioral data in Spotify is similar to the workforce matching technology used in Crowdworks! Based on the behavioral data of 360,000 crowd workers and their work history on the platform, Crowdworks recommends the most suitable workforce for the project and conducts the project through the process of selecting workers who match the customer's conditions. Due to this unique HR matching technology, Crowdworks enables the most efficient processing of data with an optimal workforce. If you're interested in Crowdworks' HR matching technology that reminds you of Spotify's playlist recommendations, please feel free to contact us!
If you have any questions about AI data, feel free to visit Crowdworks!
AI For Everyone, Crowdworks.
