%0 Journal Article
%T 采用词向量注意力机制的双路卷积神经网络句子分类模型
%A 郭宝震
%A 左万利
%A 王英
%J 浙江大学学报(工学版)
%D 2018
%R 10.3785/j.issn.1008-973X.2018.09.013
%X 针对句子中不同的词对分类结果影响不同以及每个词对应的词向量受限于单一词向量训练模型的特点，提出一种基于词向量注意力机制的双路卷积神经网络句子分类模型（AT-DouCNN）.该模型将注意力机制和卷积神经网络相结合，以不同训练算法得到的词向量同时作为输入，分别进行卷积和池化，并在全连接层进行融合，不仅能够使得具体分类任务下句子中的关键信息更易被提取，还能够有效地利用不同种类的词向量得到更加丰富的句子特征，进而提高分类的准确率.实验结果表明：所提出的模型在3个公开数据集上的分类准确率分别达到50.6%、88.6%和95.4%，具有良好的句子分类效果.</br>Abstract: A novel sentence classification model was proposed based on double convolutional neural networks with attention mechanism of word embeddings (AT-DouCNN) in view of the points that different words have different influences to the results of classification and the word embedding of each word is restricted by a single training tool. The proposed model combined the convolutional neural networks with attention mechanism. Meanwhile, this model took the word embeddings obtained by different training algorithms as input, performed convolution and pooling respectively, and fused them in the full connection layer. Based on these, the model not only makes the key information in a sentence more easily extracted under a specific classification task, but also gets more abundant sentence features with the effective use of different kinds of word embeddings, so as to improve the accuracy of classification. The experimental results demonstrate that the proposed model achieves competitive performance in sentence classification and the accuracy is 50.6%, 88.6% and 95.4% on three public datasets, respectively.
%U http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2018.09.013