计算机应用研究2025，Vol.42Issue(6)：1742-1748,7.DOI:10.19734/j.issn.1001-3695.2024.11.0442
基于句子转换和双注意力机制的归纳关系预测
Inductive relation prediction based on sentence Transformer and dual attention mechanism
摘要
Abstract
Relation prediction is a important task in knowledge graph completion,aimed at predicting missing relationships between entities.Existing inductive relation prediction methods often face challenges in adequately modeling semantic and structural information.To address this issue,this paper proposed an inductive relation prediction model based on sentence transformation and a dual-attention mechanism.The proposed method enhanced entity semantic representations by automatical-ly retrieving descriptions and incorporates a dual-attention mechanism,which considered edge and relation awareness,to accu-rately model the complex interactions between entities.Firstly,it extracted the closed subgraph of the target triple and used a random walk strategy to search for multi-hop relational paths.These triples and paths were then transformed into natural lan-guage sentences,generating semantically rich sentence embeddings.Next,it updated the subgraph embeddings using GCN and bidirectional GRU,combining sentence and subgraph embeddings to capture both structural and semantic information.Ex-perimental results on three public datasets—WN18RR,FB15k-237,and NELL-995—demonstrate that the proposed method outperforms existing methods in both transformation and inductive relation prediction tasks,validating the importance of the dual-attention mechanism and sentence transformation in improving model performance.This approach effectively enhances the accuracy and efficiency of relation prediction in knowledge graphs.关键词
知识图谱/归纳关系预测/句子转换/双注意力机制/随机行走寻径策略Key words
knowledge graph(KG)/inductive relation prediction/sentence Transformer/dual attention mechanism/random walk pathfinding strategy分类
信息技术与安全科学引用本文复制引用
李卫军,刘雪洋,刘世侠,王子怡,丁建平,苏易礌..基于句子转换和双注意力机制的归纳关系预测[J].计算机应用研究,2025,42(6):1742-1748,7.基金项目
宁夏高等学校科学研究项目(NYG2024086) （NYG2024086）
宁夏自然科学基金资助项目(2021AAC03215) （2021AAC03215）
中央高校科研资助项目(2022PT_S04,2021JCYJ12) （2022PT_S04,2021JCYJ12）
国家自然科学基金资助项目(62066038,61962001) （62066038,61962001）
北方民族大学研究生创新项目(YCX24127) （YCX24127）
