Case report: Quantitative recognition of virtual human technology acceptance based on efficient deep neural network algorithm

• Published in: Frontiers in neurorobotics Vol. 16; p. 1009093
• Main Authors: Wang, Xu, Chen, Charles
• Format: Journal Article
• Language: English
• Published: Lausanne Frontiers Research Foundation 26.10.2022; Frontiers Media S.A
• Subjects: acceptance; Artificial intelligence; Case reports; Deep learning; digital virtual human; neural network algorithm; Neural networks; Neuroscience; Robotics; Secondary schools; Students; Teaching; UTAUT model; China
• ISSN: 1662-5218; 1662-5218
• DOI: 10.3389/fnbot.2022.1009093

With the advancement of artificial intelligence, robotics education has been a significant way to enhance students' digital competency. In turn, the willingness of teachers to embrace robotics education is related to the effectiveness of robotics education implementation and the sustainability of robotics education. Two hundred and sixty-nine teachers who participated in the “virtual human education in primary and secondary schools in Guangdong and Henan” and the questionnaire were used as the subjects of study. UTAUT model and its corresponding scale were modified by deep learning algorithms to investigate and analyze teachers' acceptance of robotics education in four dimensions: performance expectations, effort expectations, community influence and enabling conditions. Findings show that 53.68% of the teachers were progressively exposed to robotics education in the last three years, which is related to the context of the rise of robotics education in schooling in recent years, where contributing conditions have a direct and significant impact on teachers' acceptance of robotics education. The correlation coefficients between teacher performance expectations, effort expectations, community influence, and enabling conditions and acceptance were 0.290 ( p = 0.000<0.001), −0.144 ( p = 0.048<0.05), 0.396 ( p = 0.000<0.001), and 0.422 ( p = 0.000<0.001) respectively, indicating that these four core dimensions both had a significant effect on acceptance. Optimization comparison results of deep learning models show that mDAE and AmDAE provide a substantial reduction in training time compared to existing noise-reducing autoencoder models. It is shown that time-complexity of the deep neural network algorithm is positively related to the number of layers of the model.