Spotify toys with deep learning to improve recommendations

Spotify could join the likes of Google and Microsoft as it looks toward deep learning to improve its recommendations.

While its music recommendation technology already uses intelligent machine learning algorithms built by Spotify Tech Lead Erik Bernhardsson, an academic researcher working from Spotify’s New York office has been researching deep learning to pull up songs based on their content.

According to a blog post written by researcher Sander Dieleman, his work aims to surface little known and popular tracks using neural networks, which take snippets of songs to create playlists based on certain features in the audio.

“I believe that music recommendation from audio signals is a pretty complex problem bridging many levels of abstraction,” said Dieleman in the post. “My hope was that successive layers of the network would learn progressively more complex and invariant features, as they do for image classification problems.”

Venture Beat has noted Facebook, Netflix and Twitter have dabbled with deep learning via various acquisitions and staff hires in the past, and Google currently uses Chinese tech company Baidu to run complex algorithms for music, news, pictures, video, and speech recognition.

Dieleman who is a fan of metal music, said the goal is to widen Spotify’s genre scope and music palate.

“I hope that this will help lesser known and up-and-coming bands, and that it will level the playing field somewhat by enabling Spotify to recommend their music to the right audience,” he said.