Here’s a fact – AI use case(s) have now become an inextricable part of our lives. From Alexa playing our favorite song to google maps finding the best route for us, AI is now a routine technology. With its multiplying potential and scope, AI and ML have made their mark in almost all the industries we can think of. And today I am here to discuss an AI use case that is making disruptions like never before. We’ll be discussing Spotify. Yes! But why the buzz around it? It’s just another online music application that gives us access to millions of songs and that’s all. Right? Well, not exactly! For starters, Spotify is indeed an online music service application, but it’s also the largest among all. It’s in the spotlight because Spotify has embraced the recent technological advancements. This AI use case has phenomenally adopted artificial intelligence and machine learning applications to enhance the customer experience.
AI Use Case: Spotify has data ( a lot of it ) but so what?
With around 248 million active users on the application, you can imagine the amount of data that the app accumulates. This data ranges from music preferences, keyword preferences to playlists data and geographical location of the users. In addition to all this data, it leverages data from several data science companies (that it has acquired). So, basically, Spotify’s success is driven by its accumulated data. All this data is utilized to train AI and ML algorithms. Spotify essentially banks on insights from the content present on the platform and also on online conversations related to music/artists and data produced by consumers to learn about its customers and in turn enhance the customer experience. So, what exactly happens when Spotify leverages data? Something like ‘Discover weekly’ happens.
The trend reached 40 million people in the first year itself!
Here’s what it is all about –
Every Monday individual users receive a customized list on thirty songs on their Spotify accounts. This playlist is essentially based on the past user behavior and the music listening/searching pattern the user has been following. The best part of machine learning is that more time = more experience = improved music recommendations.
The top two benefits of Discover Weekly are –
- The users receive a new playlist that may contain songs that they haven’t heard before so they get to explore more.
- Whereas the artists and creators are also benefitted as they are displayed on users’ playlist organically.
Behind the scenes of Discover Weekly
Here are the three models (used in combination) to create AI use case- Discover Weekly on Spotify-
- Collaborative filtering
Spotify banks on implicit feedback. It has data about the following –
- Tracks you listen regularly.
- Tracks you save to your playlist.
- Artists’ pages you visited after listening to a song.
So Spotify now has a basic picture of what your music personality looks like and what your tastes/preferences are. Now that Spotify has the data, here’s what it does through collaborative filtering- It matches your music personality with other people. And people with similar tastes are recommended each other’s preferred music. So, suppose there are two people – you and I. We both like 5 songs each and Spotify finds that 4 of them are the same. It then recommends my 5th song to you and your 5th song to me. Having said that, it is obviously not that simple. Because among millions of users the whole process is bound to be more complicated than this. But at least you know what it looks on a basic level.
2. Natural Language Processing (NLP):
Natural language processing is through which machine understand our (human) language. Here’s how Spotify leverages it – AI-powered Spotify scans – track’s metadata, blog posts, discussions about musicians, news articles on songs and singers on the internet. This is what powers Spotify to know what exactly the world is talking about the song and its artists. Through all this, it picks up descriptive terms, phrases, and other such texts associated with songs/artists.
Now that it (Spotify) knows what songs and artists the world is talking about and what language it is using to do the same, the keywords it picks are put under the umbrella of ‘cultural vectors’ and ‘top terms’. Finally, the songs and their artists are associated with thousands of these top terms (not constant). Also, to recognize which terms are more important than the other, specific weights are assigned to these terms (reflecting the number of times individuals would associate the terms with their favorite songs/artists). This also enables the app to identify the latest/trending music terms (in English and other languages derived from Latin).
3. Audio models:
Along with collaborative filtering and NLP, Spotify uses the third technique to bring you a refined discover weekly playlist. This is the audio model. Under this, all audio tracks are analyzed and songs are then classified as per their personality. For example, songs that have a high tempo, are acoustic, or are more energetic – are put together in one group. This model uses the convolutional neural network to analyze the songs. These neural networks work with high efficiency and accuracy to categorize the songs in the right groups.
But why is the audio model needed and how is it different from the aforementioned models?
See, the first model – collaborative filtering ensures that the songs are liked by a group of people so those songs are already popular. NLP too ensures the same as it displays those songs about which we talk on social media and on the internet in general. The audio model, on the other hand, gives new songs/artists a chance to rise. As it doesn’t matter if many people like the song or the song are popular or not, the audio model will analyze it nonetheless. This gives new artists the chance to have their songs on discover weekly. The aforementioned displays a fraction of what AI use case(s) are capable of. What we saw was the basic overview of how music recommendations on Spotify work. The details are far more fascinating.
And thus the right time to learn about it and leverage the opportunities it brings is – right now! Here’s a resource to help you with the same – Springboard’s courses on data science, data analytics, and Artificial Intelligence/Machine Learning are 1:1 mentoring-led and project-driven programs. The best part is they are industry focussed and comes with a job gurantee especially designed for technology enthusiasts, like you, to serve them with a career that matters.