SXSW 2017 – Thursday 16th March

Twisting television: music videos and machine learning

Some knowledgeable experts in the room. Refreshing as I was losing hope given the past 2 days. Brought me back to my digital music days 10 years ago!

  • Bouke Huurnink, Tech at XITE
  • Mickael Addicott, curation Pandora
  • Paul Natale, music curation at XITE
  • Dexter Thomas, Vice news

Challenge in curation is the long tail, less known artists given that in past 20 years it’s been easier to release or publish music or music videos. Curation is based on mood taste profiles, fine balance to push new discoveries, not too much to overpower user. Lesson learned from radio and TV.

We all have a feeling of the DJ inside us.

Curation is still human based with musicologists, music experts. AI, machine learning brings scale.

Human genome / user profile helps bring 2M unique playlists in the US at one time. Next frontier: context, to serve music based on who you are with, time of day, mood… without asking customer.

Feedback loop: monitor listeners signals e.g. Missing content yet to be released, blogs, ideas for new types of music… ‘Next big sound’ acquisition ‘Yi’ get social signals on less known music.

Conclusion: Mobilise people to monitor trends + bring machines to bring data perspective. Weekly curated playlists: watch tracks that never gets skipped. Pandora hired people with library information science background: they define scoring e.g. Cultural attributes for hip-hop + measure standard deviations + make adjustments as culture is always evolving.


Ended the festival by touring the VR corner which was hugely disappointing given the hype around VR…