Mini magick20170414 20710 katfi1

Hitwizard: Predicting hit songs with Artificial Intelligence

238 days ago

Share on

Why do some songs dominate the charts for months and most others quietly slip back into obscurity? Groningen tech companies Goldmund, Wyldebeast & Wunderliebe and Radio Radar developed an algorithm that uses machine learning to predict tomorrow’s hit songs. It’s still in beta, but the guys behind Hitwizard were already invited to pitch at Eurosonic/Noorderslag and the SXSW festival in Austin, Texas.

Around 15,000 different tracks are played weekly on the Dutch radio, but only a handful of those songs will become hits. Based on historical data, Hitwizard searches for connections and correlations that are rather difficult for people to identify, but still clear enough to be modeled by a computer. Combined with the A.I. principle of machine learning, the algorithm also learns from all that data. So the more data it has to work with, the more accurate it becomes in its predictions.

A truckload of data

Ard Boer is a serial music entrepreneur and one of the Art Directors at web agency Goldmund Wyldebeast & Wunderliebe. Radio Radar, one of his other companies, is responsible for all the music data needed for Hitwizard: “Radio Radar monitors songs and commercials played on Dutch radio and television, so artists, composers, labels and brands can track it, which makes things like paying out royalties a lot easier. So we basically had a database of over 2 million songs to work with.”

Together with Zakarias Nordfält-Laws, a developer at Goldmund, Wyldebeast & Wunderliebe, Ard developed the Hitwizard algorithm in about 3 months. “It initially just started out as a cool exercise that just got a little out of hand”, Ard continues. “Zakarias was really interested in Artificial Intelligence and I just had all this data, so basically one thing led to another.”

Defining a hit

Music, and especially taste in music, is of course very subjective. “Music is emotionally charged, so it’s really interesting to try to predict what sort of factors contribute to making a song a popular song”, Ard explains. “Even something like shitty weather could play a role in that. In the end, for people it’s really more of a gut feeling, so figuring out parameters for an algorithm is a really interesting challenge.”

All the song data from Radar Radio alone is not enough to make predictions. For extra song characteristics, Hitwizard gets a little help from Spotify. “They use their own algorithms for their playlists and suggestions, so with their help, we also know the key and tempo of a song, and if it’s energetic. Our definition of a hit is when a song ends up in the top 200 most played tracks in Spotify.”

“In total, we used 15 different parameters to analyze 85.000 tracks between June and November 2016”, Ard continues. “Hitwizard was able to predict hits in 56% of the cases, but was able to predict songs not becoming hits 99% of the time, so that’s already pretty impressive! And over time, the algorithm will only become better at learning to spot new connections. We’re still in beta of course, so we’re still busy in figuring out what kind of song characteristics to add to help our algorithm learn.”

Kicking it at SXSW

The potential Hitwizard has for the music industry hasn’t gone unnoticed. The guys were invited to speak at a seminar for music industry bigwigs at the Eurosonic/Noorderslag festival, and we’re also part of the New Dutch Wave delegation at South by Southwest (SXSW) last month, the biggest music, tech & entertainment showcase festival in the world.

Speaking at SXSW was really an amazing experience”, Ard says. “Hell, just being there is really awesome, especially if you’re a creative person. From cool bands to seeing the cast of Veep, to listening to Japanese robots discuss which is better: ramen or sushi.”

“But I thought there were far too few people from Groningen there”, he continues. “If we say we’re a digital city, or the digital capital of Europe, we should be represented by a lot of people there too. We’re barely scratching the surface of what’s possible with A.I., and I think Groningen is a very good place to start experimenting and even make it one of our main themes.”

“So, I want to see more folks from Groningen there next year! But I also want to feel that I’m in the US meeting new people, so I don’t want to talk to them”, he adds, laughing.

Check out:

https://goldmund-wyldebeast-wunderliebe.nl/

https://radio-radar.nl/ 


237 days ago

Share on