传统的知识图谱表示学习模型主要聚焦于三元组内部的结构信息,而未能充分利用外部语义增强嵌入表征能力,如没有充分考虑实体间的多步关系路径信息以及不同路径的重要程度,且没有利用实体描述信息增强上下文感知能力。为提升知识图谱的...传统的知识图谱表示学习模型主要聚焦于三元组内部的结构信息,而未能充分利用外部语义增强嵌入表征能力,如没有充分考虑实体间的多步关系路径信息以及不同路径的重要程度,且没有利用实体描述信息增强上下文感知能力。为提升知识图谱的应用效果,提出融合多步关系路径和实体描述信息的知识图谱表示学习(MPDRL)模型。首先,对两实体间的路径信息进行编码,并使用自注意力机制计算路径权重,从而获得关系路径信息的表示;其次,使用BERT(Bidirectional Encoder Representations from Transformers)模型对实体描述信息进行编码,并利用双向注意力机制计算实体描述信息嵌入与三元组关系嵌入之间的注意力权重,从而增强实体的语义信息;最后,将关系路径信息嵌入、实体描述信息嵌入和三元组结构嵌入融合起来进行训练。为评估模型性能,在公开数据集上针对所提模型和基准模型进行链接预测和三元组分类的实验。结果表明:在链接预测任务中,与融合关系路径与实体描述信息的知识图谱表示学习方法(PDRL)、多跳关系路径模型Att-ConvBiLSTM以及融合实体描述与关系路径信息的知识图谱嵌入模型TPKGE相比,所提模型在FB15k-237数据集上的Hit@10指标分别提高了5.7、2.9、2.5个百分点;在三元组分类任务上,所提模型在FB15k-237和WN18RR数据集上的准确率较最优基准模型PDRL分别提升了2.81和0.90个百分点。展开更多
Background There is a growing focus on using various plant-derived agricultural by-products to increase the benefits of pig farming,but these feedstuffs are fibrous in nature.This study investigated the relationship b...Background There is a growing focus on using various plant-derived agricultural by-products to increase the benefits of pig farming,but these feedstuffs are fibrous in nature.This study investigated the relationship between dietary fiber physicochemical properties and feedstuff fermentation characteristics and their effects on nutrient utilization,energy metabolism,and gut microbiota in growing pigs.Methods Thirty-six growing barrows(47.2±1.5 kg)were randomly allotted to 6 dietary treatments with 2 apparent viscosity levels and 3β-glucan-to-arabinoxylan ratios.In the experiment,nutrient utilization,energy metabolism,fecal microbial community,and production and absorption of short-chain fatty acid(SCFA)of pigs were investigated.In vitro digestion and fermentation models were used to compare the fermentation characteristics of feedstuffs and ileal digesta in the pig’s hindgut.Results The production dynamics of SCFA and dry matter corrected gas production of different feedstuffs during in vitro fermentation were different and closely related to the physical properties and chemical structure of the fiber.In animal experiments,increasing the dietary apparent viscosity and theβ-glucan-to-arabinoxylan ratios both increased the apparent ileal digestibility(AID),apparent total tract digestibility(ATTD),and hindgut digestibility of fiber components while decreasing the AID and ATTD of dry matter and organic matter(P<0.05).In addition,increasing dietary apparent viscosity andβ-glucan-to-arabinoxylan ratios both increased gas exchange,heat production,and protein oxidation,and decreased energy deposition(P<0.05).The dietary apparent viscosity andβ-glucanto-arabinoxylan ratios had linear interaction effects on the digestible energy,metabolizable energy,retained energy(RE),and net energy(NE)of the diets(P<0.05).At the same time,the increase of dietary apparent viscosity andβ-glucan-to-arabinoxylan ratios both increased SCFA production and absorption(P<0.05).Increasing the dietary apparent viscosity andβ-glucan-to-arabinoxylan ratios increased the diversity and abundance of bacteria(P<0.05)and the relative abundance of beneficial bacteria.Furthermore,increasing the dietaryβ-glucan-to-arabinoxylan ratios led to a linear increase in SCFA production during the in vitro fermentation of ileal digesta(P<0.001).Finally,the prediction equations for RE and NE were established.Conclusion Dietary fiber physicochemical properties alter dietary fermentation patterns and regulate nutrient utilization,energy metabolism,and pig gut microbiota composition and metabolites.展开更多
文摘传统的知识图谱表示学习模型主要聚焦于三元组内部的结构信息,而未能充分利用外部语义增强嵌入表征能力,如没有充分考虑实体间的多步关系路径信息以及不同路径的重要程度,且没有利用实体描述信息增强上下文感知能力。为提升知识图谱的应用效果,提出融合多步关系路径和实体描述信息的知识图谱表示学习(MPDRL)模型。首先,对两实体间的路径信息进行编码,并使用自注意力机制计算路径权重,从而获得关系路径信息的表示;其次,使用BERT(Bidirectional Encoder Representations from Transformers)模型对实体描述信息进行编码,并利用双向注意力机制计算实体描述信息嵌入与三元组关系嵌入之间的注意力权重,从而增强实体的语义信息;最后,将关系路径信息嵌入、实体描述信息嵌入和三元组结构嵌入融合起来进行训练。为评估模型性能,在公开数据集上针对所提模型和基准模型进行链接预测和三元组分类的实验。结果表明:在链接预测任务中,与融合关系路径与实体描述信息的知识图谱表示学习方法(PDRL)、多跳关系路径模型Att-ConvBiLSTM以及融合实体描述与关系路径信息的知识图谱嵌入模型TPKGE相比,所提模型在FB15k-237数据集上的Hit@10指标分别提高了5.7、2.9、2.5个百分点;在三元组分类任务上,所提模型在FB15k-237和WN18RR数据集上的准确率较最优基准模型PDRL分别提升了2.81和0.90个百分点。
基金supported by the National Key Research and Development Program(No.2021YFD1300201)Jilin Provincial Department of Science and Technology Innovation Platform and Talent Special Project(No.20230508090RC).
文摘Background There is a growing focus on using various plant-derived agricultural by-products to increase the benefits of pig farming,but these feedstuffs are fibrous in nature.This study investigated the relationship between dietary fiber physicochemical properties and feedstuff fermentation characteristics and their effects on nutrient utilization,energy metabolism,and gut microbiota in growing pigs.Methods Thirty-six growing barrows(47.2±1.5 kg)were randomly allotted to 6 dietary treatments with 2 apparent viscosity levels and 3β-glucan-to-arabinoxylan ratios.In the experiment,nutrient utilization,energy metabolism,fecal microbial community,and production and absorption of short-chain fatty acid(SCFA)of pigs were investigated.In vitro digestion and fermentation models were used to compare the fermentation characteristics of feedstuffs and ileal digesta in the pig’s hindgut.Results The production dynamics of SCFA and dry matter corrected gas production of different feedstuffs during in vitro fermentation were different and closely related to the physical properties and chemical structure of the fiber.In animal experiments,increasing the dietary apparent viscosity and theβ-glucan-to-arabinoxylan ratios both increased the apparent ileal digestibility(AID),apparent total tract digestibility(ATTD),and hindgut digestibility of fiber components while decreasing the AID and ATTD of dry matter and organic matter(P<0.05).In addition,increasing dietary apparent viscosity andβ-glucan-to-arabinoxylan ratios both increased gas exchange,heat production,and protein oxidation,and decreased energy deposition(P<0.05).The dietary apparent viscosity andβ-glucanto-arabinoxylan ratios had linear interaction effects on the digestible energy,metabolizable energy,retained energy(RE),and net energy(NE)of the diets(P<0.05).At the same time,the increase of dietary apparent viscosity andβ-glucan-to-arabinoxylan ratios both increased SCFA production and absorption(P<0.05).Increasing the dietary apparent viscosity andβ-glucan-to-arabinoxylan ratios increased the diversity and abundance of bacteria(P<0.05)and the relative abundance of beneficial bacteria.Furthermore,increasing the dietaryβ-glucan-to-arabinoxylan ratios led to a linear increase in SCFA production during the in vitro fermentation of ileal digesta(P<0.001).Finally,the prediction equations for RE and NE were established.Conclusion Dietary fiber physicochemical properties alter dietary fermentation patterns and regulate nutrient utilization,energy metabolism,and pig gut microbiota composition and metabolites.
