An Integrated Intelligent Teaching Evaluation System Empowered by Multimodal Large Models

Abstract

To address the structural contradiction between large-scale education and personalized assessment confronting the current "Four Evaluations" reform, and to overcome the drawbacks of traditional evaluation methods such as the separation of process and outcome, high subjectivity, and a lack of timeliness, this study proposes and designs an integrated intelligent teaching evaluation system empowered by Multimodal Large Models. The core of this system is designed around three main steps. First, it integrates multimodal data streams, including audio, vision, and text, to establish a unified data foundation for the in-depth perception and quantitative representation of all classroom elements. Second, grounded in core educational theories such as formative assessment, cognitive diagnosis, and multiple intelligences, a multidimensional evaluation indicator framework is proposed. Finally, by leveraging multi-task learning and joint inference algorithms, an integrated intelligent evaluation engine is conceptualized. This engine is capable of simultaneously producing outcome, process, value-added, and comprehensive evaluations within a unified computational framework. Theoretical analysis and architectural design suggest that this system will shift the evaluation focus from being solely score-oriented to being competency-oriented, and from summative assessment to formative empowerment. It aims to establish a real-time, closed-loop feedback mechanism that effectively integrates evaluation with teaching, thereby providing a practical technological pathway for implementing large-scale individualized instruction and fostering educational equity.