Classroom network structure learning engagement and parallel temporal attention LSTM based knowledge tracing.

Abstract

In order to accurately assess the students' learning process and the cognitive state of knowledge points in smart classroom. A classroom network structure learning engagement and parallel temporal attention LSTM based knowledge tracing model (CL-PTKT) is proposed in this paper. First, a classroom network is constructed based on the information of student ID, seating relationship, and head-up/head-down state obtained from the smart classroom video. Second, a learning engagement model is established by utilizing the head-up/head-down state of students and the structural characteristics of the classroom network. Finally, in this paper innovatively proposes a parallel temporal attention LSTM feature tracking algorithm based on the learning engagement model and the knowledge-exercise data. It can fully considers the potential associated attributes of the knowledge-knowledge, knowledge-exercise and knowledge-learning engagement. And accurately characterizes the knowledge state during the knowledge point's lecture time. To provide effective support for teachers to make seat adjustments and accurate interventions for in the teaching process. This paper conducts extensive experiments under four real datasets. The algorithm in this paper shows optimal performance in all four evaluation metrics compared to the state-of-the-art knowledge tracing models.