Abstract: To address the issuse of the insufficient collaborative modeling of semantic and structural information and the inadequate utilization of multilayer features in knowledge graph node classification with limited labels, a semi-supervised graph convolutional and attention-based jump network is proposed in this paper. The model employs a graph convolutional network to extract structural features, a graph attention network to adaptively aggregate neighborhood information, and a long short-term memory network combined with an attention mechanism to fuse node representations from different layers. Experimental results on the computer science paper citation network dataset in the open graph benchmark demonstrate that the model achieves an accuracy of 75.96%, a precision of 73.52%, a recall of 72.34%, and a harmonic mean of precision and recall of 72.93%, which outperforms several mainstream graph neural network models. The results demonstrate that the proposed model effectively integrates node semantics, graph structures, and cross-layer features, thereby improving the performance of node classification  under limited supervision.

Key words: knowledge graph, graph convolutional network, graph attention mechanism, jumping knowledge mechanism, semi-supervised learning