A Comprehensive Review on Sentiment Perception Using Electroencephalography (EEG)

Identification of sentiments using EEG is challenging and functional research area in human–computer interaction. In recent years, considerable work is done regarding detection and classification of emotions in affective computing. This review paper aims to comprehensively summarize various techniques and methods last updated in this field. The results of various techniques are mapped quantitatively by evaluating famous publications like sentiment analysis using EEG signals, detection of emotions using bio-potential signals, linear discriminant analysis (LDA) classifiers, identification of emotions from multichannel EEG through deep forest. It delineates an integrated informative approach in which various aspects of statistics from continuum of structured data sources are placed together. The most recent publications are inspected to examine the reliable approach for detection of sentiments. Moreover, there are some specific inputs for each research which are helpful to improve the performance of existing approaches in practical applications. The analysis and comparison of all the methods show that identification of emotions using multi-channel EEG through deep forest, facial expression recognition and recognition of sentiments using classifiers including LDA and SVM show best accuracies than state-of-art methods. We analyzed results of standard signals to measure the rare artifact-eliminated EEG signals. DEAP and DREAMER are challenging datasets which are being used in most of the techniques for detection and analysis of sentiments. Other datasets like GAPED, MANHOB HCl, ACSERTAINL, MULSEMEDIA and DECAF are also analyzed in various methods of sentiment analysis. The main target of this survey is to provide nearly full image of techniques regarding analysis of sentiments (detection of emotions and building resources). It is observed that the traditional procedure of feature extraction is followed along with addition of some new features in recent publications. Among all methods, it is observed that deep forest model for analysis of sentiments is oblivious to hyper-parameter settings that lead reducing complexity of recognition of sentiments.