多标签文本分类是自然语言处理领域的重要任务之一.文本的标签语义信息与文本的文档内容有紧密的联系,而传统的多标签文本分类方法存在忽略标签的语义信息以及标签的语义信息不足等问题.针对以上问题,提出一种融合标签嵌入和知识感知的...多标签文本分类是自然语言处理领域的重要任务之一.文本的标签语义信息与文本的文档内容有紧密的联系,而传统的多标签文本分类方法存在忽略标签的语义信息以及标签的语义信息不足等问题.针对以上问题,提出一种融合标签嵌入和知识感知的多标签文本分类方法 LEKA (Label Embedding and Knowledge-Aware).该方法依赖于文档文本以及相应的多个标签,通过标签嵌入来获取与标签相关的注意力.考虑标签的语义信息,建立标签与文档内容的联系,将标签应用到文本分类中.另外,为了增强标签的语义信息,通过知识图谱嵌入引入外部感知知识,对标签文本进行语义扩展.在AAPD和RCV1-V2公开数据集上与其他分类模型进行了对比,实验结果表明,与LCFA (Label Combination and Fusion of Attentions)模型相比,LEKA的F1分别提高了3.5%和2.1%.展开更多
Single crystals of CeMn_(0.85)Sb_(2) have been successfully synthesized by using the Bi as flux.Analysis of single crystal x-ray diffraction data confirms that CeMn_(0.85)Sb_(2) crystallizes in the HfCuSi_(2)-type str...Single crystals of CeMn_(0.85)Sb_(2) have been successfully synthesized by using the Bi as flux.Analysis of single crystal x-ray diffraction data confirms that CeMn_(0.85)Sb_(2) crystallizes in the HfCuSi_(2)-type structure with the space group P4/nmm(No.129).In the case of H‖c,CeMn_(0.85)Sb_(2) displays a robust antiferromagnetic transition at~160 K for Mn-sublattice,and there is no sign of magnetic order regarding Ce-sublattice.In the case of the Mn-sublattice shows signs of magnetic order at 160 K and 116 K,indicating a possible spin reorientation.There is no sign of magnetic order for the Cesublattice either,but,alternating current magnetic susceptibility measurements reveal a spin glass state below 18 K in the case of H⊥c.Isothermal magnetization curves measured below magnetic order with H⊥c show saturation and even large hysteresis at 2 K,indicating the presence of a ferromagnetic component.In addition,a field-induced spin-flop transition is observed in the case of H⊥c,indicating a field-induced spin reorientation of Mn spins.Electrical resistivity measurements indicate a metallic nature for CeMn_(0.85)Sb_(2) and large anisotropy which is consistent with its quasi-two-dimensional layered structure.展开更多
文摘多标签文本分类是自然语言处理领域的重要任务之一.文本的标签语义信息与文本的文档内容有紧密的联系,而传统的多标签文本分类方法存在忽略标签的语义信息以及标签的语义信息不足等问题.针对以上问题,提出一种融合标签嵌入和知识感知的多标签文本分类方法 LEKA (Label Embedding and Knowledge-Aware).该方法依赖于文档文本以及相应的多个标签,通过标签嵌入来获取与标签相关的注意力.考虑标签的语义信息,建立标签与文档内容的联系,将标签应用到文本分类中.另外,为了增强标签的语义信息,通过知识图谱嵌入引入外部感知知识,对标签文本进行语义扩展.在AAPD和RCV1-V2公开数据集上与其他分类模型进行了对比,实验结果表明,与LCFA (Label Combination and Fusion of Attentions)模型相比,LEKA的F1分别提高了3.5%和2.1%.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.U22A6005,U2032204,and 12104492)the Guangdong Major Scientific Research Project(Grant No.2018KZDXM061)+3 种基金the National Key Research and Development Program of China(Grant No.2021YFA1400401)the Strategic Priority Research Program of the Chinese Academy of Sciences(Grant No.XDB33010000)the K.C.Wong Education Foundation(Grant No.GJTD-2018-01)the Informatization Plan of Chinese Academy of Sciences(Grant No.CAS-WX2021SF0102)。
文摘Single crystals of CeMn_(0.85)Sb_(2) have been successfully synthesized by using the Bi as flux.Analysis of single crystal x-ray diffraction data confirms that CeMn_(0.85)Sb_(2) crystallizes in the HfCuSi_(2)-type structure with the space group P4/nmm(No.129).In the case of H‖c,CeMn_(0.85)Sb_(2) displays a robust antiferromagnetic transition at~160 K for Mn-sublattice,and there is no sign of magnetic order regarding Ce-sublattice.In the case of the Mn-sublattice shows signs of magnetic order at 160 K and 116 K,indicating a possible spin reorientation.There is no sign of magnetic order for the Cesublattice either,but,alternating current magnetic susceptibility measurements reveal a spin glass state below 18 K in the case of H⊥c.Isothermal magnetization curves measured below magnetic order with H⊥c show saturation and even large hysteresis at 2 K,indicating the presence of a ferromagnetic component.In addition,a field-induced spin-flop transition is observed in the case of H⊥c,indicating a field-induced spin reorientation of Mn spins.Electrical resistivity measurements indicate a metallic nature for CeMn_(0.85)Sb_(2) and large anisotropy which is consistent with its quasi-two-dimensional layered structure.
