Transformer-LLM Smart Classroom Assessment System: Sequence Labeling and Multi-Task Learning Methods Oriented to Ideological and Political Education in Curriculum

This article builds a scheme which uses Large Language Modeling (LLM) to push forward the combination of smart classroom teaching assessment and ideological and political education. According to a three-step method-”teaching language identification-ideology and politics component picking-integration road creating”-a two-road semantic model frame is constructed. Through the fusing of multi-head attention mechanism and multi-task loss function, the united model building and end-to-end deduction of classroom behaviors and value semantic meanings are obtained. This system has completed training and validation through utilizing the THUCTC-EDU2023 corpus, hence it obtains better performance than mainstream baseline methods on both semantic classification tasks and ideological and political education recognition tasks. It keeps steady exactness and recall on cross-course transfer learning and gets affirmative responses from teachers and students in actual teaching situations concerning evaluation correctness, suggestion matching, and interaction delay. The platform is placed and run by using Python 3.10/PyTorch and NVIDIA A100, therefore it builds a full-link mechanism which includes API scheduling (Flask), result management (MySQL), and visual playing back. This study gives a repeatable calculation frame and expandable system realization for the intelligent cooperation of”teaching” and”education.”