Development of a Computer-Based Evaluation System for Teaching Quality Using Deep Neural Networks and Optimization Algorithms

In the design of the evaluation system, Deep Neural Network (DNN) model is employed as the core component. The model first standardizes the evaluation data through the preprocessing layer of the evaluation model, followed by feature extraction and selection from the standardized data. To enhance the accuracy and efficiency of the evaluation, this paper further introduces an optimization algorithm layer, which achieves performance improvement through DNN parameter adjustment, model selection, and regularization techniques. In the experimental simulation section, this paper compares the performance of the traditional RNN, LSTM evaluation models, and the improved GUR teaching quality evaluation model. The results show that the system incorporating the GUR model and AdamW optimization algorithm demonstrates superior performance in evaluation metrics such as F1 score and matching rate, effectively enhancing the accuracy and efficiency of evaluations. These findings clearly highlight the GUR model's superiority and generalization ability in handling related tasks. Additionally, a comparative analysis of RNN, LSTM, and the proposed GUR evaluation model is conducted across process evaluation indicators, technical means evaluation, and teaching system transfer rate and response time. Evidently, the GUR model outperforms the other two models in key information capture, data collection volume and accuracy, data processing efficiency, and overall performance. Moreover, the AdamW optimization algorithm plays a pivotal role in improving the model's training stability and evaluation accuracy.