A semantic driven adaptive framework for few-shot knowledge graph completion

doi:10.1016/j.neucom.2025.131763

A semantic driven adaptive framework for few-shot knowledge graph completion

2025-10-10

0

PRE

AI

DOI：10.1016/j.neucom.2025.131763

Original

Original request for help

Save

Abstract

En 中文

• We leverage LLMs to extract and generate rich semantic descriptions for entities by analyzing their neighborhood information within the knowledge graph. This approach effectively introduces useful semantic information, enhancing the model’s ability to represent entities and their background relations. • We propose a semantics-driven negative sampling framework to generate high-quality negative samples tailored to relational contexts. Combined with adaptive fine-tuning using LoRA, this approach improves model’s adaptability to unseen data. • In link prediction tasks on the NELL-One and FB15k237-One datasets, our method achieves significant results, demonstrating its effectiveness.

A semantic driven adaptive framework for few-shot knowledge graph completion

Abstract

Journal

Organization