基于SWCNN和双向LSTM的文本情感特征分析方法  

Text emotion analysis method using SWCNN and BiLSTM

作　　者：许学添[1] 赖河蒗[1] XU Xuetian;LAI Helang(Department of Information Administration,Guangdong Justice Police Vocational College,Guangzhou 510520,China)

机构地区：[1]广东司法警官职业学院信息管理系,广州510520

出　　处：《智能计算机与应用》2023年第11期275-280,共6页Intelligent Computer and Applications

基　　金：广东省普通高校特色创新项目(2020KTSCX272);广东省普通高校特色创新项目(2022KTSCX287);广东省科技创新战略专项资金项目(pdjh2021b0815)。

摘　　要：针对文本句子中语义角色重叠、高维度文本词向量训练中难以收敛等问题,将情感词标签与卷积神经网络相结合,采用结合情感词的卷积神经网络算法,将词语转为情感标签后与词向量拼接再输入卷积神经网络,将输出的特征再与双向长短期记忆神经网络所获取的特征进行融合,最后通过全连接网络输出情感分类结果。实验结果表明,在微博新冠疫情评论情绪数据集上,本研究所提出的算法模型文本情感特征识别精确度达到89.23%,比其他深度学习算法在准确率上至少提高1.95%,而且训练具有更快的收敛速度,能够为文本情感识别提供一种新的思路与方法。Aiming at problems of overlapping semantic roles in text sentences and difficult convergence in the training of highdimensional text word vectors,this paper uses a convolution neural networks algorithm which combines emotional word labels with convolution neural networks.After words are transformed into emotional labels,they are spliced with word vectors and then input into convolution neural networks,the output features are fused with the features obtained by bidi-rectional long short-term memory neural network,finally the emotion classification results are output through the full connection network.The experimental results show that on the microblog COVID-19 comment emotion data set,the accuracy of text emotion feature recognition of the algorithm model proposed in this study reaches 89.23%,which is at least 1.95%higher than other deep learning algorithms,and the training has faster convergence speed,which can effectively improve the effect of text emotion recognition and analysis.

关 键 词：文本情感特征 自然语言处理 卷积神经网络 双向长短期记忆神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]