随着自媒体的蓬勃发展,任何人都可以在网上随意发布和转发信息,而这些信息可能是真实的,也可能是道听途说或被故意篡改的。互联网上数据的严重冗余和弱可信问题,导致现有数据的可用性很差。Bi-LSTM-CRF(Bi-Long Short Term Memory with ...随着自媒体的蓬勃发展,任何人都可以在网上随意发布和转发信息,而这些信息可能是真实的,也可能是道听途说或被故意篡改的。互联网上数据的严重冗余和弱可信问题,导致现有数据的可用性很差。Bi-LSTM-CRF(Bi-Long Short Term Memory with Conditional Random Field Layer)网络虽然能够解决数据中命名实体识别的准确率问题,但不能满足识别出的实体是可信的这一要求。文中提出一种基于编辑距离的多实体可信确认算法,并通过人物命名实体识别实例对该算法进行验证。首先通过分布式爬虫抓取同一个邮箱地址在多个搜索引擎上的Top N网页记录,然后使用经过双语语料训练后的Bi-LSTM-CRF模型抽取每个页面内的人物命名实体,最后通过实体多参数融合确定邮箱所对应的人物命名实体。实验结果表明,多实体可信确认算法能够将邮箱地址与邮箱真实主人的匹配准确率MRR(Mean Reciprocal Rank)提高到91.32%,相比只使用词频的算法其MRR提升了23.08%。实验数据充分说明,多实体可信确认算法能很好地从弱可信数据中获得强可信度的实体,降低海量数据中的低质特性,从而有效地增强实体数据源的可信度。展开更多
The hybrid genetic algorithm is utilized to facilitate model parameter estimation.The tri-dimensional compression tests of soil are performed to supply experimental data for identifying nonlinear constitutive model of...The hybrid genetic algorithm is utilized to facilitate model parameter estimation.The tri-dimensional compression tests of soil are performed to supply experimental data for identifying nonlinear constitutive model of soil.In order to save computing time during parameter inversion,a new procedure to compute the calculated strains is presented by multi-linear simplification approach instead of finite element method(FEM).The real-coded hybrid genetic algorithm is developed by combining normal genetic algorithm with gradient-based optimization algorithm.The numerical and experimental results for conditioned soil are compared.The forecast strains based on identified nonlinear constitutive model of soil agree well with observed ones.The effectiveness and accuracy of proposed parameter estimation approach are validated.展开更多
文摘随着自媒体的蓬勃发展,任何人都可以在网上随意发布和转发信息,而这些信息可能是真实的,也可能是道听途说或被故意篡改的。互联网上数据的严重冗余和弱可信问题,导致现有数据的可用性很差。Bi-LSTM-CRF(Bi-Long Short Term Memory with Conditional Random Field Layer)网络虽然能够解决数据中命名实体识别的准确率问题,但不能满足识别出的实体是可信的这一要求。文中提出一种基于编辑距离的多实体可信确认算法,并通过人物命名实体识别实例对该算法进行验证。首先通过分布式爬虫抓取同一个邮箱地址在多个搜索引擎上的Top N网页记录,然后使用经过双语语料训练后的Bi-LSTM-CRF模型抽取每个页面内的人物命名实体,最后通过实体多参数融合确定邮箱所对应的人物命名实体。实验结果表明,多实体可信确认算法能够将邮箱地址与邮箱真实主人的匹配准确率MRR(Mean Reciprocal Rank)提高到91.32%,相比只使用词频的算法其MRR提升了23.08%。实验数据充分说明,多实体可信确认算法能很好地从弱可信数据中获得强可信度的实体,降低海量数据中的低质特性,从而有效地增强实体数据源的可信度。
基金Project(2007CB714006) supported by the National Basic Research Program of China Project(90815023) supported by the National Natural Science Foundation of China
文摘The hybrid genetic algorithm is utilized to facilitate model parameter estimation.The tri-dimensional compression tests of soil are performed to supply experimental data for identifying nonlinear constitutive model of soil.In order to save computing time during parameter inversion,a new procedure to compute the calculated strains is presented by multi-linear simplification approach instead of finite element method(FEM).The real-coded hybrid genetic algorithm is developed by combining normal genetic algorithm with gradient-based optimization algorithm.The numerical and experimental results for conditioned soil are compared.The forecast strains based on identified nonlinear constitutive model of soil agree well with observed ones.The effectiveness and accuracy of proposed parameter estimation approach are validated.
