Research on Knowledge Concept Recommendation Algorithm With Spatial-Temporal Information

Abstract

Online learning is an important complementary form of offline classroom learning, in order to meet the personalized learning needs of students in the online learning process and improve the effectiveness of online learning, this paper proposes a knowledge concept recommendation algorithm based on spatio-temporal information. The algorithm models students from both the spatial and temporal dimensions to accurately recommend a personalized knowledge concept learning list. Spatially, a heterogeneous information network (HIN) based on the MOOC platform is constructed, and an improved graph convolutional network TSA-GCN is used to learn the representation of students and knowledge concepts under the meta-paths, which can adaptively aggregate the information of neighboring nodes through the trainable self-weight adjacency matrix, fully taking into account the differences in the student’s perceptual ability, and using the attention mechanism to fuse the information under multiple meta-paths to ensure the information integrity. Temporally, Attention RNN (Attention RNN, ARNN) is used to learn students’ temporal learning behaviors, mine the interest offset features in the learning process, and predict students’ current learning interests. The spatial and temporal information is fed into the extended matrix factorization model to generate the final knowledge concept recommendation list. Experiments on publicly available MOOC datasets show that the method proposed in this paper can more accurately predict and recommend the knowledge concepts that students are interested in compared to the latest proposed methods.

Keywords

• Knowledge concept recommendation
• graph convolutional networks
• recurrent neural networks
• online learning