This academic article presents CADReN, a new method for estimating node importance in Knowledge Graphs (KGs), crucial for integrating external information into Large Language Models. Traditional methods lack adaptability to new graphs and user-specific requirements. CADReN addresses these limitations by introducing a Contextual Anchor mechanism, enabling the network to assess node importance considering both structural and semantic features. CADReN proves to perform better in cross-graph NIE tasks and matches the performance of previous models on single-graph NIE tasks. The authors also introduce two new datasets for cross-graph NIE research.
Publication date: 6 Feb 2024
Project Page: https://arxiv.org/abs/2402.05135v1
Paper: https://arxiv.org/pdf/2402.05135
