基于CNN-BIGRU的中文文本情感分类模型

Chinese Comment Sentiment Classification Model Based on CNN-BIGRU

在当今商业领域,对网络评论的情感分类一直是一个比较热门的研究方向,而为了克服传统机器学习方法所构建分类器会产生较大计算开销,精度表现较差的缺点,提出一种基于深度学习模型中卷积神经网络(CNN)与循环神经网络(RNN)模型的情感分类方法。在以往的研究中,卷积神经网络往往被用来提取文本的局部特征信息,但却容易忽视文本的长距离特征,而RNN则往往被用来提取句子的长距离依赖信息,但容易陷入梯度爆炸问题。因此,结合卷积神经网络对于局部特征信息的良好提取能力与循环神经网络对于长距离依赖信息的记忆能力,构建了一个CNN-BIGRU混合模型,用以提取文本的局部特征以及文本的长距离特征。其中循环神经网络模型使用了双向GRU模型,以避免RNN模型的梯度爆炸与梯度消失问题。在谭松波的酒店评论数据集上的实验结果表明,利用该模型,实验分类的准确率比单独使用卷积神经网络模型最高提升了26.3%,比单独使用循环神经网络模型最高提升了7.9%,从而提高了对中文文本情感分类的精度,并减少了计算开销。

In today’s business field,the sentiment classification of online comments has always been a hot research direction.In order to overcome the shortcomings of the classifier constructed by the traditional machine learning method,such as large computational overhead and poor accuracy,a sentiment classification method based on the convolutional neural network(CNN)and recurrent neural network(RNN)in the deep learning model is proposed.In previous studies,CNN is often used to extract the local feature information of the text,but it is easy to ignore the long-distance feature of the text,while RNN is often used to extract the long-distance dependent information of the sentence,but it is easy to fall into the gradient explosion.Therefore,combining the great local feature information extraction of CNN and the memory of RNN to long-distance dependent information,we construct a CNN-BIGRU hybrid model to extract local feature and long-distance feature of text.A two-way GRU model is used in RNN model to avoid the gradient explosion and gradient disappearance of the RNN model.The experiment on Tan Songbo’hotel reviews data set shows that the classification accuracy of the proposed model is the highest by 26.3% compared with the CNN alone,and the highest by 7.9% compared with RNN alone,so as to improve the accuracy of the affection of Chinese text classification and reduce the computational overhead.