一种基于扩散概率分布的时序知识图谱推理方法

本发明公开了一种基于扩散概率分布的时序知识图谱推理方法，包括：基于输入数据获取实体和关系的嵌入，将所述实体与时间的嵌入进行拼接，获取当前时刻的四元组嵌入；基于加噪变化对当前时刻的四元组嵌入进行向前扩散，获取趋近收敛于标准正态分布的潜在变量，计算所述潜在变量与所述四元组嵌入的条件概率分布；基于高斯采样对所述潜在变量进行反向扩散，获取下一时刻的四元组嵌入；计算损失函数，基于所述损失函数进行优化。本发明通过多个子模块联合建模显示地学习度量空间中实体语义随时间的动态表示，能够得到更为精确的建模。 The invention discloses a time sequence mapping knowl...

Full description

Saved in:
Bibliographic Details
Format Patent
LanguageChinese
Published 23.01.2024
Subjects
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
ELECTRIC DIGITAL DATA PROCESSING
PHYSICS
Online AccessGet full text

Cover

More Information
Summary:本发明公开了一种基于扩散概率分布的时序知识图谱推理方法，包括：基于输入数据获取实体和关系的嵌入，将所述实体与时间的嵌入进行拼接，获取当前时刻的四元组嵌入；基于加噪变化对当前时刻的四元组嵌入进行向前扩散，获取趋近收敛于标准正态分布的潜在变量，计算所述潜在变量与所述四元组嵌入的条件概率分布；基于高斯采样对所述潜在变量进行反向扩散，获取下一时刻的四元组嵌入；计算损失函数，基于所述损失函数进行优化。本发明通过多个子模块联合建模显示地学习度量空间中实体语义随时间的动态表示，能够得到更为精确的建模。 The invention discloses a time sequence mapping knowledge domain inference method based on diffusion probability distribution, which comprises the following steps: acquiring embedding of entities and relationships based on input data, splicing the embedding of the entities and time, and acquiring tetrad embedding at the current moment; forward diffusion is carried out on tetrad embedding at the current moment based on noise adding change, potential variables which approach to and converge to standard normal distribution are obtained, and conditional probability distribution of the potential variables and tetrad embedding is calculated; performing back diffusion on the potential variable based on Gaussian sampling to obtain tetrad embedding at the next moment; and cal
Bibliography:Application Number: CN202310379854