事件抽取是自然语言处理(Natural Language Processing,NLP)领域的一个研究热点。现有的事件抽取模型大多基于小规模训练集,无法应用于大规模开放领域。针对大规模开放域事件抽取中事件表征困难的问题,提出了一种基于Zipf’s共生矩阵分...事件抽取是自然语言处理(Natural Language Processing,NLP)领域的一个研究热点。现有的事件抽取模型大多基于小规模训练集,无法应用于大规模开放领域。针对大规模开放域事件抽取中事件表征困难的问题,提出了一种基于Zipf’s共生矩阵分解的事件向量计算方法。首先,从开放语料中提取事件元组作为事件标签,并对事件元组进行抽象、剪枝和消歧。然后,利用Zipf’s共生矩阵表示事件的上下文分布,利用主成分分析(Principal Component Analysis,PCA)对共生矩阵进行分解,得到初始事件向量,并利用自编码器对初始事件向量进行非线性变换。采用最近邻检测和事件检测两种任务对事件向量的性能进行测试,结果表明,基于Zipf’s共生矩阵分解得到的事件向量能够对事件之间的相似性和相关性信息进行全局性表征,避免编码过细而造成语义偏移。展开更多
The geochemical characteristics of saturated and aromatic hydrocarbons from different formations and lithologies provide ob-vious evidence for transgressions that occurred during Upper Triassic Xujiahe stage in Sichua...The geochemical characteristics of saturated and aromatic hydrocarbons from different formations and lithologies provide ob-vious evidence for transgressions that occurred during Upper Triassic Xujiahe stage in Sichuan Basin with a great impact on the source input and depositional environment.A clear dual peak distribution for normal alkanes and obvious abundant com-pounds sourced from bacteria and algae in whole oil gas chromatogram indicates the abundance of lower organisms input.The ratio of Pr/Ph is low,ranging from 0.33 to 0.86 with an average of 0.60,quite different from Pr/Ph >2.0 for coal measures in swamp environment,representing source rocks from saline lake or marine facies.In the gas source rocks extracts,abundant β-carotane,-carotane,and their degradated series were detected in the whole oil chromatogram,indicating a reducing envi-ronment.The concentrations of methyl steranes and dinosteranes are high.The content of polycyclic aromatic sulfur heterocy-cles(PASH) is relatively higher in aromatic fraction and the assemblage of fluorene,dibenzofuran,and dibenzothiophene is different from the typical saline lake and the regular swamp facies source rocks,manifesting the transgression effects on gas source rocks.展开更多
Studies of repetition priming have found two face-sensitive event-related potential(ERP) components:the N250 r showing positive deflection at frontal region and negative deflection at temporal region, and the N400 sho...Studies of repetition priming have found two face-sensitive event-related potential(ERP) components:the N250 r showing positive deflection at frontal region and negative deflection at temporal region, and the N400 showing positive deflection at frontal and centro-parietal regions, both of which depend in part upon the presence or absence of a pre-existing face representation. However, the N250 r is rarely reported for a repetition interval between immediate repetition and 3 min; in addition, whether different types of representations function in the same way is also of interest. The goal of the present experiment is to compare the ERP patterns for faces versus letter strings as a function of the pre-existing memory representation with a repetition interval of 1.5 min on average. We found reliable frontally positive N250 r and N400 for famous faces and words; marginally significant effects for pseudo-words;and only the centro-parietal N400 for unfamiliar faces.Collectively, the N250 r persists in the present intermediate intervals, and both the frontal N250 r and the frontal N400 are domain-general, sensitive to the pre-existing memory representation.展开更多
文摘事件抽取是自然语言处理(Natural Language Processing,NLP)领域的一个研究热点。现有的事件抽取模型大多基于小规模训练集,无法应用于大规模开放领域。针对大规模开放域事件抽取中事件表征困难的问题,提出了一种基于Zipf’s共生矩阵分解的事件向量计算方法。首先,从开放语料中提取事件元组作为事件标签,并对事件元组进行抽象、剪枝和消歧。然后,利用Zipf’s共生矩阵表示事件的上下文分布,利用主成分分析(Principal Component Analysis,PCA)对共生矩阵进行分解,得到初始事件向量,并利用自编码器对初始事件向量进行非线性变换。采用最近邻检测和事件检测两种任务对事件向量的性能进行测试,结果表明,基于Zipf’s共生矩阵分解得到的事件向量能够对事件之间的相似性和相关性信息进行全局性表征,避免编码过细而造成语义偏移。
基金supported by National Science and Technology Major Pro-jects(Grant No.2008ZX05007-001)National Natural Science Foun-dation of China (Grant No.40973041)
文摘The geochemical characteristics of saturated and aromatic hydrocarbons from different formations and lithologies provide ob-vious evidence for transgressions that occurred during Upper Triassic Xujiahe stage in Sichuan Basin with a great impact on the source input and depositional environment.A clear dual peak distribution for normal alkanes and obvious abundant com-pounds sourced from bacteria and algae in whole oil gas chromatogram indicates the abundance of lower organisms input.The ratio of Pr/Ph is low,ranging from 0.33 to 0.86 with an average of 0.60,quite different from Pr/Ph >2.0 for coal measures in swamp environment,representing source rocks from saline lake or marine facies.In the gas source rocks extracts,abundant β-carotane,-carotane,and their degradated series were detected in the whole oil chromatogram,indicating a reducing envi-ronment.The concentrations of methyl steranes and dinosteranes are high.The content of polycyclic aromatic sulfur heterocy-cles(PASH) is relatively higher in aromatic fraction and the assemblage of fluorene,dibenzofuran,and dibenzothiophene is different from the typical saline lake and the regular swamp facies source rocks,manifesting the transgression effects on gas source rocks.
基金the National Natural Science Foundation of China (31300831)Zhejiang Provincial Social Science Foundation of China (14NDJC012Z)
文摘Studies of repetition priming have found two face-sensitive event-related potential(ERP) components:the N250 r showing positive deflection at frontal region and negative deflection at temporal region, and the N400 showing positive deflection at frontal and centro-parietal regions, both of which depend in part upon the presence or absence of a pre-existing face representation. However, the N250 r is rarely reported for a repetition interval between immediate repetition and 3 min; in addition, whether different types of representations function in the same way is also of interest. The goal of the present experiment is to compare the ERP patterns for faces versus letter strings as a function of the pre-existing memory representation with a repetition interval of 1.5 min on average. We found reliable frontally positive N250 r and N400 for famous faces and words; marginally significant effects for pseudo-words;and only the centro-parietal N400 for unfamiliar faces.Collectively, the N250 r persists in the present intermediate intervals, and both the frontal N250 r and the frontal N400 are domain-general, sensitive to the pre-existing memory representation.
