A New Method of Teaching Implementation for Large-Model Agent Evaluation: Use of Instructional Design to Achieve Degree

Abstract

The development of large model agents and chain-of-thought reasoning solves the challenges in text evaluation. The consistency of teaching content has long been an important part of teacher evaluation, and the consistency based solely on national standards does not take into account the personal development of teachers, and the use of instructional design as a mediating condition to intervene in the optimization of teachers' teaching content has become a more effective method. In this study, the DeepseekR1 large model was used to replace the manual assessment, and a method was developed to evaluate the achievement of instructional design using intelligent agents. There are five parts that can be identified in an instructional design text: instructional objectives, key difficulties, instructional process, instructional strategies, and assessment. The intelligent agent tool was used for text utterance analysis, and a new calculation method was developed based on semantic matching and achievement range, which was assigned weight (ω) to enhance its accuracy and stability. The model is fine-tuned in this way. The results showed that there were significant differences between novice teachers and expert teachers in the achievement of different parts of instructional design.