advertisement

Follow Mint Lounge

Latest Issue

Home > News> Big Story > Your streaming app knows which song you will play next

Your streaming app knows which song you will play next

Product and design experts at Spotify explain how algorithms look for ‘signals’ from users

Mood, it turns out, is a key factor not only for users but also for music streaming services.

A recent study by global data research firm Nielsen and music-streaming service Spotify offers insights into the musical preferences of Indians. The Culture & Audio Streaming Trends found users in the 13-17 age group are more likely to listen to music for a change of mood and to pass time, than the average Indian listener. Users in the 25-29 age group, more than any other age category, are more likely to listen to tracks to keep themselves motivated and stay productive. The findings are based on data collected earlier this year from 20,000 respondents across the country.

Mood, it turns out, is a key factor not only for users but also for music streaming services. There is an intricate link between technology and mood that comes alive whenever you are plugged in. For a June 2019 study published in Réseaux, a French journal of reference for social science research in the field of communication, researchers and sociologists studied the effect of algorithmic recommendation tools in a music streaming app on some 4,000 users who were selected randomly.

“The socio-technical environment of streaming puts the listener in a specific position, in many ways different from that of the listener of records or downloaded music,” the study notes. As the sociologists pointed out, these recommendation and guidance tools include everything from a collaborative filtering tool (people who liked this song also liked that one), themed recommendations (other songs by the artist you listened to) and radio stations (featuring music inspired by this artist) to playlists “inspired by your tastes”, and “atmosphere”-based playlists tailored to user moods.

Owen Smith, senior director of product at Spotify, says that when the digital music service introduced the “Browse” tab a couple of years ago, it was based on the insight that users were curating music in a very different way from record stores. There are more than four billion playlists on the platform and users curate everything from workout and sleep music to content one listens to while cooking, he says.

Users in the 25-29 age group, more than any other age category, are more likely to listen to tracks to keep themselves motivated and stay productive

“It’s actually just the way that we as humans think about music. A lot of people think about music for a particular mood... For you, your sleep music might be different to my sleep music. A lot of people listen to rain sounds, piano music. Some people listen to hip hop to work out. Mood is really important...as a concept for how users think about music, and that is reflected in how we curate it for them,” says Smith in a video call.

No wonder that in the last few months of the pandemic, playlists with keywords related to “at-home” activities have been trending globally—people have been creating everything from, “coloring-themed” playlists to music for “gardening”. Hobby podcasts have been trending, too, along with home school and working-from-home themed playlists. There were some 2,750 Spotify playlists dedicated to banana bread alone.

How the algorithm kicks in

So how does the app personalize music for users? Smith says: “If you do listen to a workout playlist at a particular time, that’s a good signal for us to maybe recommend that playlist at a similar time to you on another day. It can seem that we know your mood but if you are listening to it every morning to work out, it’s a signal for us.”

According to the IMI-IFPI Digital Music Study 2019, conducted across nine locations in India, users spent around 19.1 hours listening to music every week and for almost 97% of them smartphones were the device of choice—music consumption through audio streaming rose across all age groups, with a 10% increase among young listeners (6-24 years) and an 8% increase among older listeners (45-54 years).

Language is a major factor that influences user selections.
Language is a major factor that influences user selections.

With the likes of YouTube Music, Amazon Prime Music, JioSaavn, Gaana, Hungama Music, Airtel Wynk, Google Play Music, Apple Music and others also present in the Indian market, there’s no dearth of free, music-streaming offerings either. Getting enough people into the premium offerings, however, is a challenge. “Paid streaming remains an opportunity of growth for the industry, with only 61% respondents consuming music through the premium tiers on offer by the digital platforms,” the study adds. An even more interesting finding of the study was regarding the role of film-linked music in driving up consumption. Vintage Bollywood tracks was another section that attracted a considerable degree of user engagement.

Smith says Bollywood and Tollywood are huge in India. But for the recommendation algorithm to get things spot on, one of the key aspects was to understand the “notion of the artist in India”. “You have the notion of an artist as a singer, and the artist as the actor who is sometimes miming the song and the artist as a composer—that doesn’t exist in that many other genres of music,”he adds. “So getting the data around the artist to help us understand and match recommendations was really important.”

For the app, another important bit of data was to understand the languages a user speaks and listens to their music in. These cues were matched with the language of performance of the tracks for more accurate recommendations. This was done through language onboarding—a feature users encounter as soon as they get into the app.

For the recommendation algorithm to get things spot on, one of the key aspects was to understand the “notion of the artist in India”

The art of prediction

Sometimes, album art comes into play as well. Smith explains how the Spotify team built a tool to scan album art for visual cues when they were facing a shortage of data on devotional tracks while building the music catalogue for India.

“We didn’t actually do any acoustic analysis, but looked at whether the album or track art had the image of a god on it and that was a really good heuristic to indicate whether this was devotional music or not,” he says. “This allowed us to tag the data properly so the algorithms would work in a proper way.”

Nicole Burrow, senior director of design for Spotify’s consumer-facing products, explains how the algorithm kicks in as soon as a user signs up, enters their email and reaches the language cluster section. With the help of location data, and other small bits of information, there is enough for the algorithm to start working with. “There are a couple of bits and pieces we collect on Day 1 that can give us a high-level general understanding of the user,” says Burrow. Besides language preferences, the app’s “taste onboarding” feature allows users to select their preferred artists and genres. This helps in building personalized playlists.

Apart from these features, a user also gives away certain “explicit” and “implicit” signals that influence the app’s content recommendations. “An explicit signal is something like a ‘skip’—if you are skipping for a few songs, we can assume that maybe we gave you the wrong recommendation in that ‘Daily Mix’ playlist,” says Burrow. Users can also hide a song they don’t like in the playlist.

But Spotify keeps a lookout for implicit signals too.

“It’s through their usage of the app that we start understanding what they do and don’t like. For example, listening to a lot of a certain artist over a long period or genre, certain types of music,” she adds. “We start gathering such implicit signals too. In the end, it’s a combination of all these things that allows us to get an idea of what we should and should not recommend to them.”

Next Story