A Knowledge Point Labeling Method by Introduce Knowledge Points Labels Information

Manual annotation of knowledge points is a high cost and labor-intensive task, while automatic annotation of knowledge points is a challenging task. The labels information of knowledge points can guide model annotation, but most of the existing automatic labeling methods ignore the labels information of knowledge points. By introducing knowledge point labels information, this paper proposes an end-to-end knowledge point annotation framework. The proposed framework is composed of feature extraction module and knowledge points annotation module introducing the labels information. We introduce and capture the information between labels based on graph convolutional networks (GCN), and combine Skip Connection to accelerate model convergence. We evaluated the proposed framework on a real-world dataset labeled by four high school courses, 29,813 sample exam questions, and 97 knowledge points. The proposed model achieves similar or better performance (2% to 15% improvement in Micro F1) than the most advanced knowledge point annotation model.