In order to improve the accuracy, continuity and interpretation ability of teaching effectiveness evaluation under the background of digital transformation of higher education, this paper proposed a multi-dimensional dynamic evaluation model construction and application method. Based on the logs of learning management platform, classroom interaction records, homework performance, stage test results and text feedback, the multi-source teaching data were standardized, feature extraction and correlation modeling, and clustering analysis and machine learning classification were combined to identify teaching status, so as to alleviate the deviation caused by static, single and unbalanced samples in traditional evaluation. On this basis, the evaluation index system was constructed around teaching design, process implementation, learning participation, result achievement and feedback support. The fuzzy comprehensive evaluation and dynamic weight adjustment mechanism were used to realize the phased measurement and continuous update of teaching effectiveness. The application results showed that the comprehensive scores of course A and course B increased from 71.8 points and 70.9 points to 87.4 points and 85.3 points respectively, and the single evaluation time was controlled at 1.84 minutes and 1.91 minutes respectively, which was significantly lower than 3.96-4.37 minutes of the control method. The evaluation errors are 4.2% and 4.5%, respectively, which are also significantly lower than the control methods. The results show that the method in this paper can clearly reflect the change of teaching status, and has better performance in evaluation efficiency, stability and reliability of results, which can provide computational support for curriculum diagnosis, teaching improvement and digital teaching governance in colleges and universities.
