基于胶囊网络和高斯自注意力的用户评论情感分析  

User review′s sentiment analysis based on capsule network and Gaussian self-attention

作　　者：纪明宇[1] 赵雪峰 贾国庆 李勃毅 JI Mingyu;ZHAO Xuefeng;JIA Guoqing;LI Boyi(School of Information&Computer Engineering,Northeast Forestry University,Harbin 150040,China)

机构地区：[1]东北林业大学信息与计算机工程学院,黑龙江哈尔滨150040

出　　处：《现代电子技术》2021年第7期95-100,共6页Modern Electronics Technique

基　　金：国家自然科学基金青年科学基金项目(61806049):基于用户标签和主题兴趣的社会媒体信息推荐研究。

摘　　要：针对传统卷积神经网络和循环神经网络在文本情感分析领域对文本特征提取存在的语义丢失、无法识别文本关键词等问题进行改进,提出一种结合高斯自注意力机制的双通道文本特征提取模型。首先,该模型利用改进的胶囊网络和词性词向量提取更深层次的文本特征;然后,使用双向长短期记忆网络提取双向的语义依赖;最后,加入高斯自注意力层得到输入信息对分类结果的注意力分布。模型在两个标准数据集上进行验证,与传统的用户评论情感分析模型相比,在准确率和F1值上均取得了明显的提升。A two-channel text feature extraction model combined with Gaussian self-attention mechanism is proposed to cope with the difficulties in the text feature extraction that semantic loss occurs and text keywords are unrecognized when the traditional convolutional neural network and recurrent neural network are applied in the field of text sentiment analysis.In the model,an improved capsule network and word property word vector are used to extract deeper text features first,and then a bi-directional long short-term memory(Bi-LSTM)is used to extract a bi-directional semantic dependency,which is added into Gaussian self-attention layer to obtain the attention distribution about the input information to the classification results.It is verified with two standard data sets that the accuracy and F1 value of the proposed model have been significantly improved in comparison with those of the traditional user review sentiment analysis model.

关 键 词：情感分析 高斯自注意力机制 文本特征提取 胶囊网络 双通道模型 实验分析 

分 类 号：TN99-34[电子电信—信号与信息处理] TP391[电子电信—信息与通信工程]