Exploring multi-granularity contextual semantics for fully inductive knowledge graph completion

Jingchao Wang, Weimin Li, Alex Munyole Luvembe, Xiao Yu, Xinyi Zhang, Fangyu Liu, Fangfang Liu, Hao Wang, Zhenhai Wang, Qun Jin

Abstract: Highlights•Propose a multi-granular contextual semantic model for fully inductive KGC.•Design two path transformation strategies to enhance PLMs’ understanding of context.•PMN can filter out unreliable paths using reasoning patterns from similar cases.•SMN can extract rich high-order contextual semantics from complete subgraphs.