科技创新与应用2024，Vol.14Issue(16)：1-6,6.DOI:10.19981/j.CN23-1581/G3.2024.16.001
基于聚类方法的负样本生成器研究与实现
摘要
Abstract
The goal of knowledge graph embedding is to generate low-dimensional continuous feature vectors for entities and relationships in knowledge graph,so that computers can mine the potential semantics of knowledge through mathematical operations.It is applied to downstream tasks such as triple completion,entity classification and entity parsing.The translation model(Trans)is a simple and effective method for embedding knowledge graph,which uses negative sampling method to improve the accuracy of knowledge graph embedding.However,the traditional negative sampling method uses random negative sampling,which is easy to generate low-quality negative triple,which leads to inaccurate training of embedded vectors of entities and relations.To solve this problem,a Negative Sampling of Similar Entities(NSSE)based on Canopy and K-means method is proposed to generate high quality negative samples.The experimental results show that the translation model using NSSE achieves better results in embedding vector generation than the original model.关键词
知识图谱嵌入/翻译模型/负采样/相似实体/聚类方法Key words
knowledge graph embedding/translation model/negative sampling/similar entities/clustering algorithm分类
信息技术与安全科学引用本文复制引用
温欣,丁怡泽,屈克将,丁建新,王海涛,王建华,王天..基于聚类方法的负样本生成器研究与实现[J].科技创新与应用,2024,14(16):1-6,6.基金项目
国家重点研发计划资助(2019YFC0312003) （2019YFC0312003）
