大规模多标签文本分类(XMTC)是从一个庞大且复杂的标签集合中查找与文本样本最相关标签的一项具有挑战性的任务。目前,基于Transformer模型的深度学习方法在XMTC上取得了巨大的成功。然而,现有方法都没能充分利用Transformer模型的优势...大规模多标签文本分类(XMTC)是从一个庞大且复杂的标签集合中查找与文本样本最相关标签的一项具有挑战性的任务。目前,基于Transformer模型的深度学习方法在XMTC上取得了巨大的成功。然而,现有方法都没能充分利用Transformer模型的优势,忽略了文本不同粒度下细微的局部语义信息,同时标签与文本之间的潜在关联尚未得到稳健的建立与利用。对此,提出了一种基于语义特征与关联注意力的大规模多标签文本分类模型SemFA(An Extreme Multi-Label Text Classification Model Based on Semantic Features and Association-Attention)。在SemFA中,首先拼接多层编码器顶层输出作为全局特征。其次,结合卷积神经网络从多层编码器浅层向量中获取局部特征。综合丰富的全局信息和不同粒度下细微的局部信息获得更丰富、更准确的语义特征。最后,通过关联注意力机制建立标签特征与文本特征之间的潜在关联,引入关联损失作为潜在关联不断优化模型。在Eurlex-4K和Wiki10-31K两个公开数据集上的实验结果表明,SemFA优于大多数现有的XMTC模型,能有效地融合语义特征与关联注意力,提升整体的分类性能。展开更多
.GN algorithm has high classification accuracy on community detection, but its time complexity is too high. In large scale network, the algorithm is lack of practical values. This paper puts forward an improved GN alg....GN algorithm has high classification accuracy on community detection, but its time complexity is too high. In large scale network, the algorithm is lack of practical values. This paper puts forward an improved GN algorithm. The algorithm firstly get the network center nodes set, then use the shortest paths between center nodes and other nodes to calculate the edge betweenness, and then use incremental module degree as the algorithm terminates standard. Experiments show that, the new algorithm not only ensures accuracy of network community division, but also greatly reduced the time complexity, and improves the efficiency of community division.展开更多
Wireless sensor networks （WSNs） can be used to collect surrounding data by multi-hop. As sensor networks have the constrained and not rechargeable energy resource, energy efficiency is an important design issue for ...Wireless sensor networks （WSNs） can be used to collect surrounding data by multi-hop. As sensor networks have the constrained and not rechargeable energy resource, energy efficiency is an important design issue for its topology. In this paper, the energy consumption issue under the different topology is studied. We derive the exact mathematical expression of energy consumption for the fiat and clustering scheme, respectively. Then the energy consumptions of different schemes are compared. By the comparison, multi-level clustering scheme is more energy efficient in large scale networks. Simulation results demonstrate that our analysis is correct from the view of prolonging the large-scale network lifetime and achieving more power reductions.展开更多
文摘大规模多标签文本分类(XMTC)是从一个庞大且复杂的标签集合中查找与文本样本最相关标签的一项具有挑战性的任务。目前,基于Transformer模型的深度学习方法在XMTC上取得了巨大的成功。然而,现有方法都没能充分利用Transformer模型的优势,忽略了文本不同粒度下细微的局部语义信息,同时标签与文本之间的潜在关联尚未得到稳健的建立与利用。对此,提出了一种基于语义特征与关联注意力的大规模多标签文本分类模型SemFA(An Extreme Multi-Label Text Classification Model Based on Semantic Features and Association-Attention)。在SemFA中,首先拼接多层编码器顶层输出作为全局特征。其次,结合卷积神经网络从多层编码器浅层向量中获取局部特征。综合丰富的全局信息和不同粒度下细微的局部信息获得更丰富、更准确的语义特征。最后,通过关联注意力机制建立标签特征与文本特征之间的潜在关联,引入关联损失作为潜在关联不断优化模型。在Eurlex-4K和Wiki10-31K两个公开数据集上的实验结果表明,SemFA优于大多数现有的XMTC模型,能有效地融合语义特征与关联注意力,提升整体的分类性能。
文摘.GN algorithm has high classification accuracy on community detection, but its time complexity is too high. In large scale network, the algorithm is lack of practical values. This paper puts forward an improved GN algorithm. The algorithm firstly get the network center nodes set, then use the shortest paths between center nodes and other nodes to calculate the edge betweenness, and then use incremental module degree as the algorithm terminates standard. Experiments show that, the new algorithm not only ensures accuracy of network community division, but also greatly reduced the time complexity, and improves the efficiency of community division.
文摘Wireless sensor networks （WSNs） can be used to collect surrounding data by multi-hop. As sensor networks have the constrained and not rechargeable energy resource, energy efficiency is an important design issue for its topology. In this paper, the energy consumption issue under the different topology is studied. We derive the exact mathematical expression of energy consumption for the fiat and clustering scheme, respectively. Then the energy consumptions of different schemes are compared. By the comparison, multi-level clustering scheme is more energy efficient in large scale networks. Simulation results demonstrate that our analysis is correct from the view of prolonging the large-scale network lifetime and achieving more power reductions.
