Lattice-based node classification method and system for incomplete heterogeneous graph

• Main Authors: PAN LI, YANG CHENGLEI, MEI GUANGXU, LIU SHIJUN
• Format: Patent
• Language: Chinese; English
• Published: 23.06.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a lattice-based node classification method and system for an incomplete heterogeneous graph, and the method comprises the steps: obtaining the incomplete heterogeneous graph which comprises nodes, edge types, the number and related attribute fields; calculating different feature sets and structure sets according to the obtained data; calculating a feature partial order set and a structure partial order set according to the obtained partial order relationship between different elements in the feature set and the structure set; according to the obtained feature partial sequence set and the structure partial sequence set, constructing a feature lattice and a structure lattice; and inputting each unit in the feature lattice and the structural lattice into the lattice-based double-attention network, and outputting a predicted value of the node pair to be predicted. 本发明提供了一种针对不完整异质图的基于格的节点分类方法及系统，所述方案包括：获取不完整异质图，其中，所述不完整异质图包括节点、边类型、数目以及相关属性字段；根据获得的数据，计算不同的特征集和结构集；根据获得的特征集和结构集中的不同元素间的偏序关系，计算出特