MINDFUL SPACE IN SENTENCES - A DATASET OF VIRTUAL EMOTIONS FOR NATURAL LANGUAGE CLASSIFICATION

Abstract Spatial emotions have played a critical role in visual-spatial environmental assessment, which can be assessed using bio-sensors and language description. However, information on virtual spatial emotion assessment with objective emotion labels and natural language processing (NLP) is insufficient in literature. Thus, designers’ ability to assess spatial design quantitatively and cost effectively is limited before the design is finalized. This research measures the emotions expressed using electroencephalograms (EEGs) and descriptions in virtual reality (VR) spaces with different parameters. First, 26 subjects experienced 10 designed virtual spaces with a VR headset (Quest 2 device) corresponding to the different space parameters of shape, height, width, and length. Simultaneously, the EEG measured the emotions of the subjects using four electrodes and the five brain waves. Second, two labels – calm and active – were produced using EEGs to describe these virtual reality spaces. Last, this labeled emotion dataset compared the differences among the virtual spaces, human feelings, and the language description of the participants in the VR spatial experience. Experimental results show that the parameter changes of VR spaces can arouse significant fluctuations in the five brain waves. The EEG brain wave signals, in turn, can label the virtual rooms with calm and active emotions. Specifically, in terms of VR spaces and emotions, the experiments find that more relative spatial height results in less active emotions, while round spaces arouse calmness in the human brain waves. Moreover, the precise connection among VR spaces, brain waves in emotion, and languages still needs further research. This research attempts to offer a useful emotion measurement tool in virtual architectural design and description using EEGs. This research identifies potentials for future applications combining physiological metrics and AI methods, i.e., machine learning for synthetic design generation and evaluation.