Brain-Computer Interfaces for Music Recommendation

We explore the opportunity to harness electroencephalograph (EEG) signals for the purpose of music recommendation. The core idea lies on the hypothesis that cortical signals captured by off-the-shelf electrodes carry enough information about mental state of a listener and can be used to build preference models over musical taste for each individual user. We present a reinforcement learning algorithm that aims to build such models over a period of time and then use it effectively to provide recommendations. Our experiments on real users indicate that the recommendation policy learnt via the brain-computer interface provides better recommendations than commercial services such as Pandora.