Research and Analysis of Emotion Recognition Systems Based on Brain-Computer Interfaces

Electroencephalography (EEG) presents an approach toward practical emotion recognition by being noninvasive, providing a rich source with precise timing, and showing resilience to interference. This paper aims to present recent developments of emotion recognition systems based on brain-computer interfaces (BCI) using EEG, starting with a view into approaches used in modeling emotional states along with the methodologies used in acquiring, collecting, and preprocessing the datasets, then exploring different features or techniques from classical time/spectral representations as well as more modern differential entropy methods including connectivity-based measurement followed by advanced deep learning methodologies such as graph neural networks, transformers, etc., and gives special attention on studies involving the fusion of multistreaming data, mainly combining EEG signals with other streams of input data contributing into the robustness and generalization of current emotion recognition models. A number of reflections are presented from setting up these studies where problems related to the heterogeneity across subjects, noise due to insufficiently defined labels or ground-truths, and low natural context validity of collected datasets occur, and is followed by the discussion and elaborated review of methodologies aiming towards providing an overview that can help building better accurate and efficient methods with ability to process raw brainwave measurements and extract insights leading to characterization of affective state changes.