ProCom: Progressive Multi-modal Knowledge Graph Completion via Adaptive Function

Xiaoyu Kang, Zhixin Shi, Degang Sun, Tengfan Weng, Liyue Ren

Abstract: The primary goal of the knowledge graph completion task is to anticipate and complete the gaps in the entities and relationships. Despite the fact that representation learning has enabled existing methods to achieve remarkable results, the majority of them employ a “one-shot” completion strategy that fails to completely account for the complexity and diversity of the completion task. The accuracy and robustness of the completion results are restricted by the fact that this method fails to account for the variations in completion difficulty among different relationships and entities. This paper suggests an innovative progressive multi-modal knowledge graph completion method, ProCom, that is inspired by the concept of progressive learning. In particular, our approach implements a progressive completion strategy that progresses from straightforward to intricate through adaptive thresholds. The ProCom has demonstrated its superior performance by achieving significant performance improvements on the DB15K and FB15K datasets after multiple cycles of completion.

External IDs:dblp:conf/icic/KangSSWR25