Analysis of ongoing EEG elicited by natural music stimuli using nonnegative tensor factorization

This study proposes a six-step approach to analyze ongoing EEG elicited by 512-second long natural music (modern tango). The spectrogram of the ongoing EEG was first produced, and then a fourth-order tensor including the spectrograms of multiple channels of multiple participants was decomposed via nonnegative tensor factorization (NTF) into four factors, including temporal, spectral and spatial components, and multi-domain features of all participants. We found one extracted temporal component by NTF significantly (p < 0.01) correlated with the temporal course of a long-term music feature, ‘fluctuation centroid’ moreover, the power of posterior alpha activity was found to be associated with this temporal component. Hence, it looks promising to apply the proposed method for analyzing other ongoing EEG elicited by other natural stimuli.