诈骗电话案件频频发生并威胁着人们的生活,含有诈骗语义的语句与内容的前后文、语句序列、局部相关内容以及显著关键词语相关.本文提出神经网络模型PEAGCNN(Position Embedding and Attention are introduced into BiGRU and CNN)对诈...诈骗电话案件频频发生并威胁着人们的生活,含有诈骗语义的语句与内容的前后文、语句序列、局部相关内容以及显著关键词语相关.本文提出神经网络模型PEAGCNN(Position Embedding and Attention are introduced into BiGRU and CNN)对诈骗电话文本分类.首先构建相关数据集,词嵌入用于表示文本,不同频率的正弦、余弦函数对文本位置信息编码并融入词嵌入向量,然后分别利用BiGRU(Bidirectional Gated Recurrent Unit)和CNN(Convolutional Neural Network)提取文本上下文相关信息、语句序列以及局部相关性,Attention机制对提取出的信息重新分配权重,突出关键信息的作用,最后将两种信息融合,通过Softmax实现分类.在THUCNews数据集以及诈骗电话文本数据集上的实验结果表明,本文提出模型的准确率和F1值均比对比模型有提升,同时模型对诈骗电话文本数据集分类的各项性能指标均在0.91以上.展开更多
电话诈骗日益猖獗,严重影响人民的生命和财产安全,如何有效防范电话诈骗已成为社会的一大焦点问题.本文提出一种基于Attention-BiLSTM模型的诈骗电话识别方法.该方法以电话文本为数据集,采用双向长短时记忆神经网络(bi-directional long...电话诈骗日益猖獗,严重影响人民的生命和财产安全,如何有效防范电话诈骗已成为社会的一大焦点问题.本文提出一种基于Attention-BiLSTM模型的诈骗电话识别方法.该方法以电话文本为数据集,采用双向长短时记忆神经网络(bi-directional long short-term memory)模型提取句子的长距离特征.通过引入注意力机制增强电话文本中与诈骗相关词汇的特征权重,得到电话文本的句子层面的特征向量表示,最后输入Softmax层进行分类预测.实验结果表明,基于注意力机制的BiLSTM诈骗电话分类模型的准确率较基线模型分别提高了2.15%和0.6%,具有更好的预测性能.展开更多
如今,电话诈骗案件层出不穷,严重危害到了人们的财产安全和社会的和谐安定。针对社会中的一些诈骗电话问题,提出了一种基于词嵌入和混合神经网络的文本分类方法,实现对诈骗电话文本的分类。首先构造了诈骗电话文本数据集,内容涵盖了金...如今,电话诈骗案件层出不穷,严重危害到了人们的财产安全和社会的和谐安定。针对社会中的一些诈骗电话问题,提出了一种基于词嵌入和混合神经网络的文本分类方法,实现对诈骗电话文本的分类。首先构造了诈骗电话文本数据集,内容涵盖了金融、教育、邮递、银行等多类诈骗事件。为了优化文本的输入词向量,词嵌入部分采用基于Transformer的BERT(Bidirectional Encoder Representation from Transformers)模型来表示诈骗文本,同时采用基于双向长短时记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)以及多尺度卷积神经网络(Convolutional Neural Network,CNN)的混合神经网络(BiLCNN)对文本的词嵌入表示进行特征提取,充分提取出文本的时序特征和局部相关特征,最后将特征融合在一起通过Softmax进行分类。通过实验比较了Word2vec、ELMo(Embedding from Language Model)和BERT三种词嵌入模型,表明BERT作为输入向量的优越性,同时在诈骗电话文本数据集上的实验结果表明,提出的模型BERT+BiLCNN相比Word2vec+CNN、ELMo+CNN和BERT+CNN模型,诈骗电话文本分类准确率分别提高了4.12%、2.84%和0.95%。展开更多
To investigate the effects of dietary supplementation with hydrolyzed wheat gluten (HWG) on growth performance, cell immunity and serum biochemical indices of weaned piglets, 160 crossed (Large White×andrace)...To investigate the effects of dietary supplementation with hydrolyzed wheat gluten (HWG) on growth performance, cell immunity and serum biochemical indices of weaned piglets, 160 crossed (Large White×andrace) and weaned piglets were randomly divided into 4 treatments with 4 replicates of 10 piglets each. The piglets in each treatment were fed an experimental diet containing either 0 g kg-1 HWG (control group), 30 g kg-1 HWG (3% HWG group), 50 g kg-1 HWG (5% HWG group), or 2.5 g kg-1 glycyl-L-glutamine (0.25% Gly-Gln group). The results showed that the diarrhea rates in 3% HWG and 5% HWG groups were significantly lower than in control group from d 1 to 14 (P〈0.05), while the average daily gain (ADG) in each of two groups was increased (P〉0.05). When compared with control group, dietary supplementation with 3% HWG increased the ratio of CD4+:CD8+ and the ratio of serum albumin and globulin concentrations (A:G) on d 14 and 28, as well as the proliferation of T- and B-lymphocytes (P〉0.05) on d 28. In addition, on d 14 and 28, the A:G ratio in 5% HWG group was significantly higher than in control group (P〈0.05), while the ratio of CD4+:CD8+ increased slightly (P〉0.05). Interestingly, 0.25% Gly-Gln group had higher proportion of CD3+ (P〉0.05) and CD4+ (P〈0.05) on d 14 than control group, but growth performances of 0.25% Gly-Gln group were negatively affected at all experiment stages. These results suggested that HWG might improve the growth performance of piglets by strengthening cell immunity and decreasing the occurrence of diarrhea during the prophase after weaning.展开更多
文摘诈骗电话案件频频发生并威胁着人们的生活,含有诈骗语义的语句与内容的前后文、语句序列、局部相关内容以及显著关键词语相关.本文提出神经网络模型PEAGCNN(Position Embedding and Attention are introduced into BiGRU and CNN)对诈骗电话文本分类.首先构建相关数据集,词嵌入用于表示文本,不同频率的正弦、余弦函数对文本位置信息编码并融入词嵌入向量,然后分别利用BiGRU(Bidirectional Gated Recurrent Unit)和CNN(Convolutional Neural Network)提取文本上下文相关信息、语句序列以及局部相关性,Attention机制对提取出的信息重新分配权重,突出关键信息的作用,最后将两种信息融合,通过Softmax实现分类.在THUCNews数据集以及诈骗电话文本数据集上的实验结果表明,本文提出模型的准确率和F1值均比对比模型有提升,同时模型对诈骗电话文本数据集分类的各项性能指标均在0.91以上.
文摘电话诈骗日益猖獗,严重影响人民的生命和财产安全,如何有效防范电话诈骗已成为社会的一大焦点问题.本文提出一种基于Attention-BiLSTM模型的诈骗电话识别方法.该方法以电话文本为数据集,采用双向长短时记忆神经网络(bi-directional long short-term memory)模型提取句子的长距离特征.通过引入注意力机制增强电话文本中与诈骗相关词汇的特征权重,得到电话文本的句子层面的特征向量表示,最后输入Softmax层进行分类预测.实验结果表明,基于注意力机制的BiLSTM诈骗电话分类模型的准确率较基线模型分别提高了2.15%和0.6%,具有更好的预测性能.
文摘如今,电话诈骗案件层出不穷,严重危害到了人们的财产安全和社会的和谐安定。针对社会中的一些诈骗电话问题,提出了一种基于词嵌入和混合神经网络的文本分类方法,实现对诈骗电话文本的分类。首先构造了诈骗电话文本数据集,内容涵盖了金融、教育、邮递、银行等多类诈骗事件。为了优化文本的输入词向量,词嵌入部分采用基于Transformer的BERT(Bidirectional Encoder Representation from Transformers)模型来表示诈骗文本,同时采用基于双向长短时记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)以及多尺度卷积神经网络(Convolutional Neural Network,CNN)的混合神经网络(BiLCNN)对文本的词嵌入表示进行特征提取,充分提取出文本的时序特征和局部相关特征,最后将特征融合在一起通过Softmax进行分类。通过实验比较了Word2vec、ELMo(Embedding from Language Model)和BERT三种词嵌入模型,表明BERT作为输入向量的优越性,同时在诈骗电话文本数据集上的实验结果表明,提出的模型BERT+BiLCNN相比Word2vec+CNN、ELMo+CNN和BERT+CNN模型,诈骗电话文本分类准确率分别提高了4.12%、2.84%和0.95%。
基金supported by the Major Special Project of Guangdong Province, China (2009A080303009)the Special Fund for Public Welfare Industry of China (Agriculture, 201003011)+2 种基金the National 948 Project of China (2011-G35)the National Major Science Research Program of China (2009CB941601)the Joint Funds of the National Natural Science Foundation of China (U0731004)
文摘To investigate the effects of dietary supplementation with hydrolyzed wheat gluten (HWG) on growth performance, cell immunity and serum biochemical indices of weaned piglets, 160 crossed (Large White×andrace) and weaned piglets were randomly divided into 4 treatments with 4 replicates of 10 piglets each. The piglets in each treatment were fed an experimental diet containing either 0 g kg-1 HWG (control group), 30 g kg-1 HWG (3% HWG group), 50 g kg-1 HWG (5% HWG group), or 2.5 g kg-1 glycyl-L-glutamine (0.25% Gly-Gln group). The results showed that the diarrhea rates in 3% HWG and 5% HWG groups were significantly lower than in control group from d 1 to 14 (P〈0.05), while the average daily gain (ADG) in each of two groups was increased (P〉0.05). When compared with control group, dietary supplementation with 3% HWG increased the ratio of CD4+:CD8+ and the ratio of serum albumin and globulin concentrations (A:G) on d 14 and 28, as well as the proliferation of T- and B-lymphocytes (P〉0.05) on d 28. In addition, on d 14 and 28, the A:G ratio in 5% HWG group was significantly higher than in control group (P〈0.05), while the ratio of CD4+:CD8+ increased slightly (P〉0.05). Interestingly, 0.25% Gly-Gln group had higher proportion of CD3+ (P〉0.05) and CD4+ (P〈0.05) on d 14 than control group, but growth performances of 0.25% Gly-Gln group were negatively affected at all experiment stages. These results suggested that HWG might improve the growth performance of piglets by strengthening cell immunity and decreasing the occurrence of diarrhea during the prophase after weaning.