Exploiting Large Language Models Capabilities for Question Answer-Driven Knowledge Graph Completion Across Static and Temporal Domains

• DOI: 10.48550/arxiv.2408.10819
• Publication Date: 2024-08-20
• Authors: Rui Yang; Jiahao Zhu; Jianping Man; Fang Li; Yixuan Zhou
• Keywords: Knowledge graph

Knowledge graph completion (KGC) aims to identify missing triples in a knowledge (KG). This is typically achieved through tasks such as link prediction and instance completion. However, these methods often focus on either static graphs (SKGs) or temporal (TKGs), addressing only within-scope triples. paper introduces new generative framework called Generative Subgraph-based KGC (GS-KGC). GS-KGC employs question-answering format directly generate target entities, the challenge of questions having multiple possible answers. We propose strategy that extracts subgraphs centered entities relationships within KG, from which negative samples neighborhood information are separately obtained address one-to-many problem. Our method generates using known facts facilitate discovery information. Furthermore, we collect refine path data providing contextual enhance reasoning large language models (LLMs). experiments evaluated proposed four SKGs two TKGs, achieving state-of-the-art Hits@1 metrics five datasets. Analysis results shows can discover existing KGs beyond closed effectively bridging gap between closed-world open-world KGC.