This article presents a new method called Anchoring Path Sentence Transformer (APST) to predict missing relations in Knowledge Graphs (KGs). The APST alleviates the reliance on Closed Paths (CPs) when many reasoning paths do not form CPs. The APST uses a search-based description retrieval method to enrich entity descriptions and an assessment mechanism to evaluate the rationality of Anchoring Paths (APs). This method takes both APs and CPs as inputs of a unified Sentence Transformer architecture, allowing comprehensive predictions and high-quality explanations. This method has been evaluated on three public datasets and has achieved state-of-the-art performance in 30 of 36 transductive, inductive, and few-shot experimental settings.
Publication date: 21 Dec 2023
Project Page: https://arxiv.org/abs/2312.13596v1
Paper: https://arxiv.org/pdf/2312.13596
