Abstract
Considering the structural characteristics and the adaptive and self-learning functions of neural networks, a teaching quality evaluation system for high-level teachers was developed using the Backpropagation (BP) neural algorithm. The system's mathematical model was established, with each evaluation index serving as input and teaching effectiveness as output. The model demonstrated feasibility and applicability in terms of convergence speed and network adaptability. Experimental results indicated that the mathematical model effectively addressed the complexity and interference of human factors in traditional analysis and evaluation of teaching processes. The model exhibited features such as convenience, accuracy, reliability, and speed, with high identification accuracy.
Introduction
Evaluating the quality of university teaching is a complex task traditionally plagued by subjectivity. Standard methods often rely on simple averages of student surveys or peer reviews, which are prone to human interference, personal bias, and the limitations of "fuzzy" logic. The core challenge lies in the non-linear nature of teaching quality; qualitative factors like "innovative concepts" interact with classroom dynamics in ways that rigid, weighted averages cannot accurately capture. This research proposes a systematic mathematical approach using neural networks to handle these non-linear relationships, aiming for a more objective and reliable evaluation framework.
Methodology
The researchers utilized a Backpropagation (BP) Neural Network, a supervised learning algorithm capable of approximating complex, non-linear functions. The model architecture consists of three distinct layers designed to map teaching behaviors to a final quality score.
- Input Layer: Comprises 22 nodes, each representing a specific secondary evaluation indicator (X1 through X22).
- Hidden Layer (Intermediate Layer): Contains 44 nodes utilizing a Sigmoid activation function to process the complex interactions between inputs.
- Output Layer: A single node representing the final teaching effectiveness score (Y).
The model was configured with a learning rate of 0.9 and an error goal of 0.001. Weights were iteratively adjusted to minimize the mean square error (MSE) between the network's predictions and the target evaluation goals.
Data and Indicator System
The study established a comprehensive hierarchy of 22 secondary indicators categorized under six primary indices:
- Teaching Idea: Innovative concepts and reform consciousness.
- Execution of Curriculum: Adherence to schedules and teaching duties.
- Teaching Content: Logical organization and introduction of new theories.
- Teaching Tactics: People-oriented methods and use of modern tools.
- Teaching Culture: Academic rigor and classroom atmosphere.
- Fundamental Teaching Skills: Language expression, board writing, and punctuality.
The dataset for training and validation consisted of 21 survey samples collected from a college. Respondents scored the 22 indicators on a scale of 0 to 9, with a corresponding "Evaluation Objective (Y)" established for each sample to serve as the ground truth for the neural network.
Results and Discussion
The experimental simulation demonstrated that the BP neural network is highly effective at identifying teaching quality with remarkable precision. For instance, in Sample 1, the target goal was 7.75, and the network predicted 7.7487. Similarly, for Sample 15, the goal was 6.0, and the network predicted 5.9994.
As shown in the convergence curve, the model reached its target error threshold of 0.001 within approximately 9 cycles (epochs). This rapid convergence indicates high computational efficiency and strong network adaptability. By automating the evaluation through a trained network, the process effectively removes the "interference of human factors" typical of manual analysis.
Conclusion and Future Work
While the model shows high identification accuracy and reliability, it is currently limited by a small sample size (21 samples), which may affect its ability to generalize across broader institutional contexts. Furthermore, while the processing model is objective, the input data remains dependent on human-scored surveys. Future research should focus on expanding the dataset to improve generalization and exploring the integration of more diverse, real-time data sources to further refine the objectivity of the evaluation process.
Summary updated