当前的基于词向量的多文档摘要方法没有考虑句子中词语的顺序,存在异句同向量问题以及在小规模训练数据上生成的摘要冗余度高的问题。针对这些问题,提出基于PV-DM(Distributed Memory Model of Paragraph Vectors)模型的多文档摘要方法...当前的基于词向量的多文档摘要方法没有考虑句子中词语的顺序,存在异句同向量问题以及在小规模训练数据上生成的摘要冗余度高的问题。针对这些问题,提出基于PV-DM(Distributed Memory Model of Paragraph Vectors)模型的多文档摘要方法。该方法首先构建单调亚模(Submodular)目标函数;然后,通过训练PV-DM模型得到句子向量计算句子间的语义相似度,进而求解单调亚模目标函数;最后,利用优化算法抽取句子生成摘要。在标准数据集Opinosis上的实验结果表明该方法优于当前主流的多文档摘要方法。展开更多
“十四五”时期,烟草行业面临社会环境发生巨大变革的挑战。从行业现状来说,烟草市场面临消费需求日益多元化、市场竞争日趋激烈、销售结构提升矛盾突出等问题。随着数字技术的不断发展,数据驱动逐渐成为烟草行业的新推手。零售户数据...“十四五”时期,烟草行业面临社会环境发生巨大变革的挑战。从行业现状来说,烟草市场面临消费需求日益多元化、市场竞争日趋激烈、销售结构提升矛盾突出等问题。随着数字技术的不断发展,数据驱动逐渐成为烟草行业的新推手。零售户数据作为最基本的数据来源,可以帮助企业有针对性地优化市场布局。本文以河南中烟CRM客户管理系统中的零售户信息为数据基础,以CRISP-DM(Cross Industry Standard Process forData Mining)为研究框架,结合逻辑回归、ARIMA时间序列、BP神经网络等机器学习和深度学习模型,对黄金叶(天叶)规格卷烟在2021年第四季度的销售数据进行建模验证,助力精准把控未来零售户价值走向。展开更多
Quantum teleportation via the entangled channel composed of a two-qubit Heisenberg XYZ model with Dzyaloshinski-Moriya （DM） interaction in the presence of intrinsic decoherence has been investigated. We find that th...Quantum teleportation via the entangled channel composed of a two-qubit Heisenberg XYZ model with Dzyaloshinski-Moriya （DM） interaction in the presence of intrinsic decoherence has been investigated. We find that the initial state of the channel plays an important role in the teleported state and the average fidelity of teleportation. When the initial channel is in the state ｜ψ1 （0）〉 = a｜00〉 ＋ b｜11〉, the average fidelity is equal to 1/3 constantly, which is independent of the DM interaction and the intrinsic decoherence effect. But when the channel is initially in the state ｜ψ2（0）〉 = a｜01〉 ＋ b｜10〉, the average fidelity is always larger than 2/3. Moreover, under a certain condition, the average fidelity can be enhanced by adjusting the DM interaction, and the intrinsic decoherence leads to a suppression of the fluctuation of the average fidelity.展开更多
文摘当前的基于词向量的多文档摘要方法没有考虑句子中词语的顺序,存在异句同向量问题以及在小规模训练数据上生成的摘要冗余度高的问题。针对这些问题,提出基于PV-DM(Distributed Memory Model of Paragraph Vectors)模型的多文档摘要方法。该方法首先构建单调亚模(Submodular)目标函数;然后,通过训练PV-DM模型得到句子向量计算句子间的语义相似度,进而求解单调亚模目标函数;最后,利用优化算法抽取句子生成摘要。在标准数据集Opinosis上的实验结果表明该方法优于当前主流的多文档摘要方法。
文摘“十四五”时期,烟草行业面临社会环境发生巨大变革的挑战。从行业现状来说,烟草市场面临消费需求日益多元化、市场竞争日趋激烈、销售结构提升矛盾突出等问题。随着数字技术的不断发展,数据驱动逐渐成为烟草行业的新推手。零售户数据作为最基本的数据来源,可以帮助企业有针对性地优化市场布局。本文以河南中烟CRM客户管理系统中的零售户信息为数据基础,以CRISP-DM(Cross Industry Standard Process forData Mining)为研究框架,结合逻辑回归、ARIMA时间序列、BP神经网络等机器学习和深度学习模型,对黄金叶(天叶)规格卷烟在2021年第四季度的销售数据进行建模验证,助力精准把控未来零售户价值走向。
基金Project supported by the National Natural Science Foundation of China (Grant Nos 60708003, 60578050 and 10434060)the National Basic Research Program of China (Grant No 2006CB921604)+1 种基金the Shanghai Science and Technology Committee (GrantNo 07JC14017)by the Director Fund of State Key Laboratory of Precision Spectroscopy
文摘Quantum teleportation via the entangled channel composed of a two-qubit Heisenberg XYZ model with Dzyaloshinski-Moriya （DM） interaction in the presence of intrinsic decoherence has been investigated. We find that the initial state of the channel plays an important role in the teleported state and the average fidelity of teleportation. When the initial channel is in the state ｜ψ1 （0）〉 = a｜00〉 ＋ b｜11〉, the average fidelity is equal to 1/3 constantly, which is independent of the DM interaction and the intrinsic decoherence effect. But when the channel is initially in the state ｜ψ2（0）〉 = a｜01〉 ＋ b｜10〉, the average fidelity is always larger than 2/3. Moreover, under a certain condition, the average fidelity can be enhanced by adjusting the DM interaction, and the intrinsic decoherence leads to a suppression of the fluctuation of the average fidelity.
文摘数据挖掘语言标准化的研究是开发新一代数据挖掘系统的关键。DMX(Data Mining Extensions,数据挖掘扩展)是OLE DBFor DM规范支持的数据挖掘查询语言,支持数据挖掘系统直接对关系数据库进行挖掘,是数据挖掘原语标准化发展中的一个突破。该文介绍了OLE DB For DM规范下数据挖掘的主要步骤,给出了Microsoft SQL Server Analysis Services中基于DMX的实现方法。
