Aiming at the deficiencies of existing learning path recommendation in knowledge dynamics, temporal correlation and scene adaptability, an adaptive learning path generation framework is proposed. Firstly, a domain KG integrating multi-dimensional knowledge attributes and dynamic relationships is constructed, and a learner cognitive state update mechanism is designed to realize the dynamic evolution of the graph. Furthermore, the learning path generation problem is modeled as a spatio-temporal sequence prediction task, where the "temporal" dimension is the learning activity sequence and the "spatial" dimension is the structure and semantics of the KG. Meanwhile, a spatio-temporal knowledge-aware network model is proposed, which uses GNN to capture the complex dependencies between knowledge nodes and combines an improved temporal convolutional network to mine the long and short-term patterns of learning behaviors, so as to achieve accurate prediction of future learning nodes and path generation. Experiments on the self-constructed dataset show that the proposed method is significantly superior to the baseline models in multiple indicators such as path effectiveness, learning efficiency gain and personalization degree.
Wang et al. (Thu,) studied this question.