摘要
考虑到电影影评上下文信息带有固有的属性特征和语序不合理性等特点,提出CRCNN模型进行文本情感分析。为了减少噪音数据对分析的影响,对卷积神经网络进行改进,在输入层和卷积层之间引入了权重分布层对重要部分进行分析,减少噪音,使处理的特征得到提升。在卷积层中使用梯度下降法来求解训练参数时会引起梯度弥散或爆炸,为了解决此问题引入了门控机制。最后引入序列标注层,同时和神经网络学习的语义特征进行整体的优化求解。另外,利用字粒度词向量为特征,解决了歧义词的切分的同时学习到更加具体的特征。实验结果表明,利用该模型进行影评分析的效果明显好于其它几种模型。
Considering the inherent characteristics of the film review context information and the irrationality of the word order,a CRCNN(Conditional Random Field Convolutional Neural Networks)model is proposed for text sentiment analysis.In order to reduce the impact of noise data on the analysis,the convolutional neural network is improved,and a weight distribution layer is introduced between the input layer and the convolutional layer to analyze the important parts,reduce the noise,and improve the processing characteristics.Using the gradient descent method in the convolutional layer to solve the training parameters will cause the gradient to spread or explode.In order to solve this problem,a gating mechanism is introduced.Finally,the sequence label layer is introduced,and the semantic features of neural network learning are optimized.In addition,the word granularity word vector is used as the feature,which solves the segmentation of the ambiguous word while learning more specific features.Experiments show that the model has significantly better film evaluation effect than other models.
作者
李俭兵
刘栗材
LI Jian-bing;LIU Li-cai(School of Telecommunication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065;Chongqing Information Technology Designing Company Limited,Chongqing 400021;Research Center of New Telecommunication Technology Applications,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
出处
《计算机工程与科学》
CSCD
北大核心
2019年第12期2261-2269,共9页
Computer Engineering & Science
关键词
影评
情感分析
卷积神经网络
序列标注层
字粒度词向量
film review
sentiment analysis
convolutional neural network
sequence labeling layer
word granularity word vector
