Evaluating Learning States in Synchronous Remote Classes via Qwen2.5-Max with RAG and ReAct Agent

In a synchronous remote class, where lectures are simultaneously delivered to both local and remote students, evaluating the group learning state of the remote class presents significant challenges. To address this problem, a series of methods for evaluating students’ learning states in the synchronous remote class based on Qwen2.5-Max is proposed. First, a behavior recognition model and a facial emotion recognition model were constructed to recognize each student’s actions and facial emotions in the class. Subsequently, Qwen2.5-Max with the RAG individual learning state recognition method is proposed. Finally, Qwen2.5-Max with ReAct agent for the synchronous remote class group learning state recognition method is proposed to determine the group learning state and provide instructional suggestions to the teacher. The proposed methods are tested using a custom-made dataset. The results indicate that the methods can help teachers balance local and remote classes by enhancing the quality of synchronous remote teaching.