Multi-Channel Graph Neural Network (MuGNN)
Patent Name: Multi-Channel Graph Neural Network (MuGNN)
Filing No: 10201906457S (SG)
Filing Date: 7 Nov 2019
Country to be Filed: SG
Description: This invention proposes a novel model Multi-Channel Graph Neural Network (MuGNN), which learns alignment-oriented embeddings by encoding graphs from different perspectives: completion and pruning, so as to be robust to structural differences. The key idea is to perform KG inference and alignment jointly, so that the heterogeneity of KGs is explicitly reconciled through completion by rule inference and transfer, and pruning via cross-KG attention, which benefits the utilization of seed alignments. Extensive experiments on five datasets and further ablation study demonstrate the effectiveness of our proposed model.