EEG-Based Evaluation of Aesthetic Experience Using BiLSTM Network
Peishan Wang,Haibei Feng,Xiaobing Du,Rui Nie,Yudi Lin,Cuixia Ma,Liang Zhang
DOI: https://doi.org/10.1080/10447318.2023.2278926
IF: 4.92
2023-01-01
International Journal of Human-Computer Interaction
Abstract:Evaluation of aesthetic design fulfills a pivotal function in product development, which urges for an efficacious objective method to measure customers' experience. The stability and effectiveness of electroencephalography (EEG) make it a suitable tool for aesthetic experience measurement. Nevertheless, existing studies have several limitations, especially regarding the stimuli and the algorithm. The potential of an EEG-based deep learning model has not been verified in pinpointing subtle differences in physical product aesthetics. To fill the research gap in this issue, we recorded EEG signals in real-life scenarios when participants were presented with different types of physical smartphones, and asked participants to rate them from four dimensions of aesthetic experience (arousal, valence, likeness, and aesthetic evaluation). Then, the time-frequency data were fed into a spatial feature extraction network and an attention-based bidirectional long short-term memory (BiLSTM) optimized by the cross-entropy loss function. The result showed that at 16s window size, the four outcome models yielded the best joint recognition performance of aesthetic experience with an average accuracy of over 85% (arousal: 88.10%, valence: 87.97%, likeness: 85.99%, and aesthetic evaluation: 87.23%). It provides an objective cross-subject recognition method with multi-faceted evaluation results of aesthetic experience. Additionally, we verified the ability of EEG as a reliable and informative resource in terms of aesthetic experience evaluation, even with subtle differences. More practically, a future direction of incorporating EEG signals into subjective product aesthetics measurement could be given more credit.