Exploration of Artificial Intelligence Technology Simulating Biological Neural Network to Optimize Teaching Evaluation System Under the Perspective of Educational Technology

Yongbin Zhou
School of Communication, Baicheng Normal University, Baicheng, Jilin 137000, China
Huaiyu Lu
Shishi Experimental School, Nanhai District, Foshan, Guangdong, 528231, China
Taoyu Chen
Shishi Experimental School, Nanhai District, Foshan, Guangdong, 528231, China

Abstract:

Aiming at the feedback delay and evaluation bias caused by insufficient data real-time and incomplete capture of individual differences in the teaching process, this study integrates the Transformer and GNN (Graph Neural Network) algorithms. Transformer efficiently extracts global dynamic features through the self-attention mechanism, and GNN accurately captures students' local interaction information. The combination of the two realizes dynamic real-time and accurate evaluation, improving feedback timeliness and model interpretability. This paper comprehensively collects multi-dimensional data in the teaching process, and uses data cleaning and normalization to ensure the quality of the original data, providing a reliable foundation for subsequent model construction. The data processing process is strictly implemented by standardization requirements. The Transformer self-attention mechanism can be used to extract global dynamic features from the collected teaching data, identify the internal connections between various links in the teaching process, focus on capturing dynamic changes in a large range, and ensure the comprehensive expression of data features. GNN is used to construct interaction graphs, accurately model local interaction information between students and between teachers and students, and refine the capture of individual behaviors and interaction characteristics. Global features are integrated with local information to construct an evaluation system that simulates biological neural networks, and multiple rounds of optimization and adjustment are performed on the model parameters. Experimental results show that in 1,000 teaching feedback time measurements, the median response time of Transformer+GNN is 120ms, Q1-Q3 is 110-130ms, and the delay is 263ms when the concurrency number is 10, which has significant advantages in maintaining the real-time nature of teaching feedback. In the three-teaching links of knowledge transfer, teacher-student interaction and teaching evaluation, the average MAE (Mean Absolute Error) of the fusion model in this paper is 0.12, and the average MSE (Mean Square Error) is 0.025, which has good prediction performance in teaching scenarios. Timely teaching feedback and good teaching scenario prediction show the optimization effect of the teaching evaluation system in this paper.