Harneshing the Potential of EEG in Neuromarketing with Deep Learning and Riemannian Geometry

Abstract

Neuromarketing exploits neuroimaging techniques to study consumers’ responses to various marketing aspects, with the goal of gaining a more thorough understanding of the decision-making process. The neuroimaging technology encountered the most in neuromarketing studies is Electroencephalography (EEG), mainly due to its non-invasiveness, low cost and portability. Opposed to typical neuromarketing practices, which rely on signal-power related features, we introduce an efficient decoding scheme that is based on the principles of Riemannian Geometry and realized by means of a suitable deep learning (DL) architecture (i.e., SPDNet). We take advantage of a recently released, multi-subject, neuromarketing dataset to train SPDNet under the close-to-real-life scenario of product selection from a supermarket leaflet and compare its performance against standard tools in EEG-based neuromarketing. The sample covariance is used as an estimator of the ‘quasi-instantaneous’, brain activation pattern and derived from the multichannel signal recorded while the subject is gazing at a given product. Pattern derivation is followed by proper re-alignment to reduce covariate shift (inter-subject variability) before SPDNet casts its binary decision (i.e., “Buy”-“NoBuy”). The proposed decoder is characterized by sufficient generalizability to derive valid predictions upon unseen brain signals. Overall, our experimental results provide clear evidence about the superiority of the DL-decoder relatively to both conventional neuromarketing and alternative Riemannian Geometry-based approaches, and further demonstrate how neuromarketing can benefit from recent advances in data-centric machine learning and the availability of relevant experimental datasets.

This work was a part of project NeuroMkt, co-financed by the European Regional Development Fund of the EU and Greek National Funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, under the call RESEARCH CREATE INNOVATE (Project code T2EDK-03661).