類似度指標のラベル付き有向グラフへの拡張
類似度指標のラベル付き有向グラフへの拡張
复制标题
将相似性度量扩展到标记有向图
DOI：
10.11517/jsaifpai.121.0_24
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
山本 章博
中科院分区：
文献类型：
--
作者：
松原 徳秀;山本 章博
分析知识图等复杂图是处理异质信息网络的一项关键任务。节点对的节点相似度是一种基本的度量，它可以应用于许多任务，如链接预测和聚类。已经提出了许多相似性度量方法，如SimRank和RoleSim。然而，包括RoleSim在内的许多相似性度量只捕获了图的结构信息，因此它们不适合于带有附加标签信息的图。我们提出了一种新的相似性度量方法LRoleSim，它同时利用图的结构信息和标签信息来度量节点之间的相似性。在我们的研究中，我们讨论了LRoleSim的性质，并证明了在一定条件下，LRoleSim具有与RoleSim相同的五个性质。此外，我们还展示了如何高效地计算LRoleSim。在真实知识图数据集上的实验验证了LRoleSim在效率和性能上优于RoleSim。
Analyzing complicated graphs, such as knowledge graphs, is a key task that handles heterogeneous information networks. Node similarity of a pair of nodes is a basic metric which can be applied to numerous tasks, eg, link prediction and clustering. Many similarity measurements, such as SimRank and RoleSim, have been proposed. However, a lot of similarity measurements, including RoleSim, capture the structural information of a graph only, so they are not suitable for graphs with additional label information. We propose a new similarity measurement called LRoleSim, which measures the similarities among nodes by using both the structural information and the label information of a graph. In our research, we discuss the properties of LRoleSim and prove that LRoleSim has the same five properties as RoleSim in certain conditions. In addition, we show how to efficiently calculate LRoleSim. Experiments on knowledge graph datasets in the real world verify that LRoleSim is superior to RoleSim in efficiency and performance.