针对词向量语义信息不完整以及文本特征抽取时的一词多义问题,提出基于BERT(Bidirectional Encoder Representation from Transformer)的两次注意力加权算法(TARE)。首先,在词向量编码阶段,通过构建Q、K、V矩阵使用自注意力机制动态编...针对词向量语义信息不完整以及文本特征抽取时的一词多义问题,提出基于BERT(Bidirectional Encoder Representation from Transformer)的两次注意力加权算法(TARE)。首先,在词向量编码阶段,通过构建Q、K、V矩阵使用自注意力机制动态编码算法,为当前词的词向量捕获文本前后词语义信息;其次,在模型输出句子级特征向量后,利用定位信息符提取全连接层对应参数,构建关系注意力矩阵;最后,运用句子级注意力机制算法为每个句子级特征向量添加不同的注意力分数,提高句子级特征的抗噪能力。实验结果表明:在NYT-10m数据集上,与基于对比学习框架的CIL(Contrastive Instance Learning)算法相比,TARE的F1值提升了4.0个百分点,按置信度降序排列后前100、200和300条数据精准率Precision@N的平均值(P@M)提升了11.3个百分点;在NYT-10d数据集上,与基于注意力机制的PCNN-ATT(Piecewise Convolutional Neural Network algorithm based on ATTention mechanism)算法相比,精准率与召回率曲线下的面积(AUC)提升了4.8个百分点,P@M值提升了2.1个百分点。在主流的远程监督关系抽取(DSER)任务中,TARE有效地提升了模型对数据特征的学习能力。展开更多
To solve the query processing correctness problem for semantic-based relational data integration,the semantics of SAPRQL（simple protocol and RDF query language） queries is defined.In the course of query rewriting,al...To solve the query processing correctness problem for semantic-based relational data integration,the semantics of SAPRQL（simple protocol and RDF query language） queries is defined.In the course of query rewriting,all relative tables are found and decomposed into minimal connectable units.Minimal connectable units are joined according to semantic queries to produce the semantically correct query plans.Algorithms for query rewriting and transforming are presented.Computational complexity of the algorithms is discussed.Under the worst case,the query decomposing algorithm can be finished in O（n2） time and the query rewriting algorithm requires O（nm） time.And the performance of the algorithms is verified by experiments,and experimental results show that when the length of query is less than 8,the query processing algorithms can provide satisfactory performance.展开更多
AIM:To examine the disparities in macular retinal vascular density between individuals with connective tissue disease-related interstitial lung disease(CTD-ILD)and healthy controls(HCs)by optical coherence tomography ...AIM:To examine the disparities in macular retinal vascular density between individuals with connective tissue disease-related interstitial lung disease(CTD-ILD)and healthy controls(HCs)by optical coherence tomography angiography(OCTA)and to investigate the changes in microvascular density in abnormal eyes.METHODS:For a retrospective case-control study,a total of 16 patients(32 eyes)diagnosed with CTD-ILD were selected as the ILD group.The 16 healthy volunteers with 32 eyes,matched in terms of age and sex with the patients,were recruited as control group.The macular retina’s superficial retinal layer(SRL)and deep retinal layer(DRL)were examined and scanned using OCTA in each individual eye.The densities of retinal microvascular(MIR),macrovascular(MAR),and total microvascular(TMI)were calculated and compared.Changes in retinal vascular density in the macular region were analyzed using three different segmentation methods:central annuli segmentation method(C1-C6),hemispheric segmentation method[uperior right(SR),superior left(SL),inferior left(IL),and inferior right(IR)],and Early Treatment Diabetic Retinopathy Study(ETDRS)methods[superior(S),inferior(I),left(L),and right(R)].The data were analyzed using Version 9.0 of GraphPad prism and Pearson analysis.RESULTS:The OCTA data demonstrated a statistically significant difference(P<0.05)in macular retinal microvessel density between the two groups.Specifically,in the SRL and DRL analyses,the ILD group exhibited significantly lower surface density of MIR and TMI compared to the HCs group(P<0.05).Furthermore,using the hemispheric segmentation method,the ILD group showed notable reductions in SL,SR,and IL in the superficial retina(P<0.05),as well as marked decreases in SL and IR in the deep retina(P<0.05).Similarly,when employing the ETDRS method,the ILD group displayed substantial drops in superficial retinal S and I(P<0.05),along with notable reductions in deep retinal L,I,and R(P<0.05).In the central annuli segmentation method,the ILD group exhibited a significant decrease in the superficial retinal C2-4 region(P<0.05),whereas the deep retina showed a notable reduction in the C3-5 region(P<0.05).Additionally,there was an observed higher positive likelihood ratio in the superficial SR region and deep MIR.Furthermore,there was a negative correlation between conjunctival vascular density and both deep and superficial retinal TMI(P<0.001).CONCLUSION:Patients with CTD-ILD exhibits a significantly higher conjunctival vascular density compared to the HCs group.Conversely,their fundus retinal microvascular density is significantly lower.Furthermore,CTD-ILD patients display notably lower superficial and deep retinal vascular density in comparison to the HCs group.The inverse correlation between conjunctival vascular density and both superficial and deep retinal TMI suggests that detecting subtle changes in ocular microcirculation could potentially serve as an early diagnostic indicator for connective tissue diseases,thereby enhancing disease management.展开更多
文摘针对词向量语义信息不完整以及文本特征抽取时的一词多义问题,提出基于BERT(Bidirectional Encoder Representation from Transformer)的两次注意力加权算法(TARE)。首先,在词向量编码阶段,通过构建Q、K、V矩阵使用自注意力机制动态编码算法,为当前词的词向量捕获文本前后词语义信息;其次,在模型输出句子级特征向量后,利用定位信息符提取全连接层对应参数,构建关系注意力矩阵;最后,运用句子级注意力机制算法为每个句子级特征向量添加不同的注意力分数,提高句子级特征的抗噪能力。实验结果表明:在NYT-10m数据集上,与基于对比学习框架的CIL(Contrastive Instance Learning)算法相比,TARE的F1值提升了4.0个百分点,按置信度降序排列后前100、200和300条数据精准率Precision@N的平均值(P@M)提升了11.3个百分点;在NYT-10d数据集上,与基于注意力机制的PCNN-ATT(Piecewise Convolutional Neural Network algorithm based on ATTention mechanism)算法相比,精准率与召回率曲线下的面积(AUC)提升了4.8个百分点,P@M值提升了2.1个百分点。在主流的远程监督关系抽取(DSER)任务中,TARE有效地提升了模型对数据特征的学习能力。
基金Weaponry Equipment Pre-Research Foundation of PLA Equipment Ministry (No. 9140A06050409JB8102)Pre-Research Foundation of PLA University of Science and Technology (No. 2009JSJ11)
文摘To solve the query processing correctness problem for semantic-based relational data integration,the semantics of SAPRQL（simple protocol and RDF query language） queries is defined.In the course of query rewriting,all relative tables are found and decomposed into minimal connectable units.Minimal connectable units are joined according to semantic queries to produce the semantically correct query plans.Algorithms for query rewriting and transforming are presented.Computational complexity of the algorithms is discussed.Under the worst case,the query decomposing algorithm can be finished in O（n2） time and the query rewriting algorithm requires O（nm） time.And the performance of the algorithms is verified by experiments,and experimental results show that when the length of query is less than 8,the query processing algorithms can provide satisfactory performance.
基金Supported by National Natural Science Foundation of China(No.82160195)Key R&D Program of Jiangxi Province(No.20223BBH80014)General Science and Technology Program of the Department of Traditional Chinese Medicine,Jiangxi Provincial Health Commission(No.2017A241).
文摘AIM:To examine the disparities in macular retinal vascular density between individuals with connective tissue disease-related interstitial lung disease(CTD-ILD)and healthy controls(HCs)by optical coherence tomography angiography(OCTA)and to investigate the changes in microvascular density in abnormal eyes.METHODS:For a retrospective case-control study,a total of 16 patients(32 eyes)diagnosed with CTD-ILD were selected as the ILD group.The 16 healthy volunteers with 32 eyes,matched in terms of age and sex with the patients,were recruited as control group.The macular retina’s superficial retinal layer(SRL)and deep retinal layer(DRL)were examined and scanned using OCTA in each individual eye.The densities of retinal microvascular(MIR),macrovascular(MAR),and total microvascular(TMI)were calculated and compared.Changes in retinal vascular density in the macular region were analyzed using three different segmentation methods:central annuli segmentation method(C1-C6),hemispheric segmentation method[uperior right(SR),superior left(SL),inferior left(IL),and inferior right(IR)],and Early Treatment Diabetic Retinopathy Study(ETDRS)methods[superior(S),inferior(I),left(L),and right(R)].The data were analyzed using Version 9.0 of GraphPad prism and Pearson analysis.RESULTS:The OCTA data demonstrated a statistically significant difference(P<0.05)in macular retinal microvessel density between the two groups.Specifically,in the SRL and DRL analyses,the ILD group exhibited significantly lower surface density of MIR and TMI compared to the HCs group(P<0.05).Furthermore,using the hemispheric segmentation method,the ILD group showed notable reductions in SL,SR,and IL in the superficial retina(P<0.05),as well as marked decreases in SL and IR in the deep retina(P<0.05).Similarly,when employing the ETDRS method,the ILD group displayed substantial drops in superficial retinal S and I(P<0.05),along with notable reductions in deep retinal L,I,and R(P<0.05).In the central annuli segmentation method,the ILD group exhibited a significant decrease in the superficial retinal C2-4 region(P<0.05),whereas the deep retina showed a notable reduction in the C3-5 region(P<0.05).Additionally,there was an observed higher positive likelihood ratio in the superficial SR region and deep MIR.Furthermore,there was a negative correlation between conjunctival vascular density and both deep and superficial retinal TMI(P<0.001).CONCLUSION:Patients with CTD-ILD exhibits a significantly higher conjunctival vascular density compared to the HCs group.Conversely,their fundus retinal microvascular density is significantly lower.Furthermore,CTD-ILD patients display notably lower superficial and deep retinal vascular density in comparison to the HCs group.The inverse correlation between conjunctival vascular density and both superficial and deep retinal TMI suggests that detecting subtle changes in ocular microcirculation could potentially serve as an early diagnostic indicator for connective tissue diseases,thereby enhancing disease management.
