字词融合的双通道混合神经网络情感分析模型

针对双向门控循环神经网络(BiGRU)无法获取文本局部特征,卷积神经网络(CNN)无法聚焦文本全局特征的问题,提出一种字词融合的双通道混合神经网络文本情感分析模型(CW_BGCA).首先,将文本分别用字符级词向量和词语级词向量表示;然后使用门控循环神经网络和卷积神经网络结合的混合神经模型分别从字向量和词向量中提取隐层特征,并分别引入注意力机制进行特征权重分配;最后将双通道网络提取的特征融合,输入到Softmax函数进行分类.在数据集上进行了多组实验验证,该方法取得了93.15%的F1值、93.47%的准确率,优于其他对照模型.试验结果表明,该模型能够有效的提高文本情感分析的性能.

Two-channel Mixed Neural Network Sentiment Analysis Model Based on Character and Word Fusion

Aiming at the problems that bidirectional gated recurrent unit(BiGRU)cannot obtain local features of text and convolutional neural network(CNN)cannot focus on global features of text,two-channel mixed neural network sentiment analysis model based on character and word fusion(CW_BGCA)is proposed.First,the text is represented by character-level vectors and word-level vectors;then,a mixed neural model combining BiGRU network and CNN network is used to extract hidden layer features from character vectors and word vectors,and introduce attention mechanisms to feature separately Weight distribution;finally,the features extracted from the two channel network are fused and input to the Softmax function for classification.Multiple experiments were performed on the data set,and the method achieved an F1 value of 93.15%and an accuracy rate of 93.47%,which is better than other control models.The experimental results show that the model can effectively improve the performance of text sentiment analysis.