In response to the common problems of static indicators (78% of universities use fixed weights) and insufficient recognition of implicit abilities (unstructured data utilization rate<12%) in the evaluation of university teachers, this study aims to construct a dynamic and accurate evaluation model to solve the problems of imbalanced subjective and objective weights and evaluation validity under small sample data.
An evaluation framework integrating combined weighting (analytic hierarchy process (AHP) + improved criteria importance through intercriteria correlation (CRITIC) method) and adaptive neural networks is proposed. By combining dynamic weighting mechanisms with adaptive feature extraction techniques, transfer learning and particle swarm optimization (PSO) are introduced to compress the training cycle.
The experimental results show that after applying this model, the interdisciplinary evaluation error decreases by 23.7%, the implicit ability recognition rate reaches 81%, the small sample (500 groups) error rate remains at 0.61%, the weight perturbation sensitivity coefficient is 0.62, the evaluation feedback delay is compressed to 1.2 days, and the graphics processing unit utilization rate increases to 89%.
For the first time, a dynamic weighted game model called “AHP Improved Critic” is established to address the bottleneck of interdisciplinary weight adaptation. Next, an attention-driven particle swarm optimized transfer learning and multi-scale VGG model (PTMVGG) network is designed to achieve a 92% feature conversion rate for unstructured data. Finally, a closed-loop system for “weight-optimized feature extraction and dynamic correction” is constructed to support the real-time generation of capability profiles.
