Knowledge Reasoning Method for Military Decision Support Knowledge Graph Mixing Rule and Graph Neural Networks Learning together

Abstract

The military decision support knowledge graphs (KGs) have been widely used in helping the commanders to understand the complicated, changefully battlefield situation. Not only combat action entities and their relations are embraced in the KGs, but also the military rules with numbers which are established by the army are encompassed. Proposed in the paer is the knowledge reasoning method mixing rule and graph neural networks learning together, called context-surrounding graph neural networks with numbers (CS-GNN-N). The rule learning, rule injection and graph neural networks learning are iteratively done in the CS-GNN-N. The two graph smoothness metrics, feature smoothness and label smoothness, are applied to measure the quantity and quality of neighborhood information of nodes respectively. Finally, the effectiveness of the CS-GNN-N on link prediction tasks is compared with four datasets and competitive reasoning methods. The results show that not only the military rules and relative numbers can be learned in the CS-GNN-N, but also the reasoning ability of the CS-GNN-N can be enhanced.