融合锚节点和三元关系向量的军事知识图谱表示学习

Military Knowledge Graph Representation Learning by Integrating Anchor Nodes and Triple Relation Vectors

知识图谱是一种用于表示和推理知识的图结构,对于军事领域的决策支持和智能化应用具有重要意义。现有的军事知识图谱多面临着大量三元组缺失的问题,对辅助决策任务造成极大的影响。为此,提出一种融合锚节点和三元关系向量的军事知识图谱表示学习模型。通过NodePiece构建固定大小的锚节点词汇表,以实现对任何实体的引导性编码和嵌入;利用TripleRE综合关系向量的投影与平移特征表示,实现对实体间语义相似度和关系强度的深层捕捉。实验结果表明,该方法在军事知识图谱单跳推理任务上取得了优异的性能,证明其有效性和可行性。

A knowledge graph is a graph structure used for representing and reasoning knowledge,which is of great significance for decision-making support and intelligent applications in the military field.However,the existing military knowledge graph mostly faces the problem of missing a large number of triples,which has a significant impact on the auxiliary decision-making task.Therefore,a military knowledge graph representation learning model that integrates anchor nodes and triple relation vectors is proposed.Firstly,a fixed-size anchor node vocabulary is constructed by NodePiece to achieve guiding coding and embedding of any entity.Secondly,the projection and translation feature representation of the TripleRE comprehensive relation vector is used to realize the deep capture of semantic similarity and relationship strength between entities.The experimental results show that the proposed method has achieved excellent performance in the single-hop reasoning task of the military knowledge graph,its effectiveness and feasibility are proved.