基于cw2vec与CNN-BiLSTM注意力模型的中文微博情感分类

Chinese Weibo Sentiment Classification Based on cw2vec and CNN-BiLSTM Attention Model

为改善中文微博领域现有中文分词不准确且微博文本不规范问题,提出将cw2vec和CNN-BiLSTM注意力相结合的混合深度学习模型方法进行中文微博情感分类。首先使用cw2vec进行中文词向量训练,转换为相应的向量作为分类模型输入,然后利用CNN提取局部特征,采用BiLSTM提取上下文特征进行深度学习,最后增加注意力模型获取重要特征,并使用Softmax函数进行分类。实验结果表明,使用cw2vec模型训练中文微博词时,准确率达到88.72%,使用cw2vec和CNN-BiLSTM注意力混合深度学习模型的精确率、召回率和F1值比单一深度学习模型分别提升1.88%、3.56%和2.72%,证明cw2vec和CNN-BiLSTM注意力混合深度学习模型结合方法在中文微博情感分类任务中的有效性。

In order to improve the problems of inaccurate Chinese word segmentation and non-standard Weibo text in the field of Chi⁃nese Weibo,a hybrid deep learning model method combining cw2vec and CNN-BiLSTM attention is proposed to carry out sentiment classification of Chinese Weibo.First,cw2vec is used to train the Chinese word vector and convert it to the corresponding vector as the input of the classification model.Secondly,CNN is used to extract local features,then BiLSTM is emploved to extract contextual fea⁃tures for deep learning,and finally the attention model is used to obtain important features and use Softmax Function for classification.The experimental results show that when using the cw2vec model to train Chinese Weibo words,the accuracy rate reached 88.72%,and the accuracy,recall rate and F1 value of the cw2vec and CNN-BiLSTM attention models have been respectively improved by 1.88%,3.56%and 2.72%compared to the single deep learning model.Experiments result show that the proposed combination of cw2vec and CNN-BiLSTM attention hybrid deep learning model is effective in the task of Chinese Weibo sentiment classification.